<?xml version="1.0" encoding="UTF-8"?>
<rss version="2.0"
     xmlns:content="http://purl.org/rss/1.0/modules/content/"
     xmlns:dc="http://purl.org/dc/elements/1.1/"
     xmlns:dcterms="http://purl.org/dc/terms/"
     xmlns:media="http://search.yahoo.com/mrss/"
     xmlns:atom="http://www.w3.org/2005/Atom"
     xmlns:cf="https://www.futureplc.com/rss/content-flags"
>
    <channel>
                    <atom:link rel="alternate" hreflang="en-GB"
                       href="https://www.itpro.com/uk/feeds/tag/ai-governance"
                       type="application/rss+xml"/>
                            <title><![CDATA[ Latest from ITPro UK in Ai-governance ]]></title>
                <link>https://www.itpro.com/uk/artificial-intelligence/ai-governance</link>
        <description><![CDATA[ All the latest ai-governance content from the ITPro  UK team ]]></description>
                                    <lastBuildDate>Tue, 01 Sep 2026 14:37:36 +0000</lastBuildDate>
                            <language>en</language>
                                <item>
                                                            <title><![CDATA[ Anthropic resumes model testing after recent cyber incidents ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Anthropic is back to testing security models after systems went rogue – and claims the incident wasn’t entirely down to security faults but AI misalignment. </p><p>Back in July, Anthropic revealed its Claude models had slipped out of their apparent bounds to hack third-party systems. Days later, its <a href="https://www.itpro.com/security/cyber-attacks/anthropics-mythos-ai-tried-to-dupe-devs-in-social-engineering-attack-collaborated-with-other-agents"><u>Mythos system was spotted</u></a> with similar alarming behavior by the UK AI Security Institute. </p><p>The incidents followed <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime"><u>OpenAI revealing similar behavior</u></a> from its own security models. </p><p>Last week, <a href="https://www.itpro.com/technology/artificial-intelligence/six-things-openai-learned-about-ai-from-the-hugging-face-incident"><u>OpenAI reported on its own incident</u></a>, detailing plans to improve security around such models, and Anthropic has followed with its own efforts, including confirmation that it will begin testing again. </p><p>Anthropic noted it was currently conducting a review of what happened and said it is "planning" to work with METR on an independent review; OpenAI has already published an independent METR review into its own incident. </p><p>But Anthropic isn't waiting to make changes. In a <a href="https://www.anthropic.com/news/improving-alignment-security-efforts" target="_blank"><u>blog post</u></a> on 31 August, the firm said it plans to make tweaks to containment and monitoring, introduce new rules for third-party testers, as well as changes to the models. </p><p>Notably, the AI developer revealed it is seconding 150 product engineers to focus primarily on security. </p><p>"We believe the incidents reflect a failure of operational security, as well as two alignment issues: motivated reasoning, and willingness to take harmful actions in pursuit of a narrow task," the company said. </p><h2 id="anthropic-is-back-testing">Anthropic is back testing</h2><p>While the full investigation isn't yet complete, Anthropic said it was removing a "pause" on external <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>evaluations of pre-release models as well as internal tests. </p><p>To avoid similar issues happening again, Anthropic is adding additional layers of security and "hardening" the sandbox where these models are set loose, noting that move was prompted by its own evidence as well as that shared by OpenAI. </p><p>The changes include a system to spot when a model attempts to escape a testing environment or obtains internet access, immediately blocking the activity and alerting people overseeing the task, as well as fixing potential flaws and misconfigurations in the sandbox. </p><p>"We migrated high-risk internal cyber sandboxes to more robust isolation," the post added. "We are also further red-teaming our virtualization stack – that is, tasking pre-release models to try and escape it – to find and patch weaknesses."</p><p>Similar protections have been applied to reinforcement learning environments, with Anthropic adding it plans to expand its monitoring across most other internal frontier agentic usage. </p><p>Beyond testing environments, Anthropic is hardening security across the company, including blocking outbound traffic by default on computing clusters, boosting identity verification on internal services, retiring legacy services, and reducing the number of human and automated accounts that can access model weights or customer data. </p><h2 id="tougher-testing-rules">Tougher testing rules</h2><p>With regard to third-party testing, Anthropic said it will require testers to commit to a fresh set of best practices, noting that the reported incidents all took place in external evaluation environments. </p><p>As <a href="https://www.itpro.com/technology/artificial-intelligence/independent-testing-firm-irregular-the-source-of-misconfigurations-that-led-to-meta-openai-and-anthropic-ai-incidents"><u><em>ITPro </em></u><u>reported last month</u></a>, Israeli startup Irregular was identified as the external model testing firm involved in all three ‘rogue AI’ incidents involving Anthropic, OpenAI, and Meta. </p><p>The changes include running tests inside hardened sandboxes with no internet access, tasking the model being tested to look for flaws before tests under close human supervision, run monitoring continuously, and confirm whether test tasks are impossible or not. </p><p>"When an evaluation target is offline or a task can’t be completed, agents will often look for other ways to complete a challenge, increasing the chance that they take actions outside the intended scope of the evaluation," the company said. </p><h2 id="unpicking-why">Unpicking why</h2><p>Those operational changes should help prevent models from escaping their sandboxes, but Anthropic said it wants to better understand why its systems took such "dangerous actions" in the first place. </p><p>While the full investigation continues, Anthropic highlighted two challenges with alignment, which refers to ensuring that AI behaves how humans want and expect. </p><p>"One is motivated reasoning: the models were initially told their environments were simulated, but when they later encountered evidence that they were connected to the real internet, they may have interpreted that evidence in a way that allowed them to maintain that belief," the blog post explained. </p><p>"The second is recklessness: the model was willing to take harmful actions on the real internet in pursuit of the narrow goal of solving a cybersecurity evaluation."</p><p>The company noted that poorly designed training environments – including ones that are easy to cheat in or impossible to solve without cheating – can lead to misalignment behavior. </p><p>Anthropic is trying to avoid such issues in the future. The company said work on thie front goes back several months and involves training poorly and training well to try to spot differences – but admitted that the July incidents show "our process isn't perfect and our models aren't perfectly aligned". </p><p>Further details on security improvements are expected in the full report, according to Anthropic. </p><p>Beyond that, the firm also loosely backed calls to develop a framework for safe development of security-focused AI, acknowledging that its own executives had recently signed an open letter demanding better coordination globally. </p><p>Anthropic said it would "say more in the coming weeks" but intended to contribute to such efforts. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/anthropic-resumes-model-testing-after-recent-cyber-incidents-but-its-introduced-new-rules-to-improve-security</link>
                                                                            <description>
                            <![CDATA[ Anthropic has boosted its security and tweaked its training to avoid rogue AI ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">prrkks6bEQDGqqZoVB8jKR</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/jLZncMSkFV4MARFdHwLLV5-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 01 Sep 2026 14:37:36 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:10 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/jLZncMSkFV4MARFdHwLLV5-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Logo of AI startup Anthropic, developer of the Claude Opus 4 and Sonnet 4 AI models, pictured on a smartphone held in a human hand.]]></media:description>                                                            <media:text><![CDATA[Logo of AI startup Anthropic, developer of the Claude Opus 4 and Sonnet 4 AI models, pictured on a smartphone held in a human hand.]]></media:text>
                                <media:title type="plain"><![CDATA[Logo of AI startup Anthropic, developer of the Claude Opus 4 and Sonnet 4 AI models, pictured on a smartphone held in a human hand.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/jLZncMSkFV4MARFdHwLLV5-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Anthropic is back to testing security models after systems went rogue – and claims the incident wasn’t entirely down to security faults but AI misalignment. </p><p>Back in July, Anthropic revealed its Claude models had slipped out of their apparent bounds to hack third-party systems. Days later, its <a href="https://www.itpro.com/security/cyber-attacks/anthropics-mythos-ai-tried-to-dupe-devs-in-social-engineering-attack-collaborated-with-other-agents"><u>Mythos system was spotted</u></a> with similar alarming behavior by the UK AI Security Institute. </p><p>The incidents followed <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime"><u>OpenAI revealing similar behavior</u></a> from its own security models. </p><p>Last week, <a href="https://www.itpro.com/technology/artificial-intelligence/six-things-openai-learned-about-ai-from-the-hugging-face-incident"><u>OpenAI reported on its own incident</u></a>, detailing plans to improve security around such models, and Anthropic has followed with its own efforts, including confirmation that it will begin testing again. </p><p>Anthropic noted it was currently conducting a review of what happened and said it is "planning" to work with METR on an independent review; OpenAI has already published an independent METR review into its own incident. </p><p>But Anthropic isn't waiting to make changes. In a <a href="https://www.anthropic.com/news/improving-alignment-security-efforts" target="_blank"><u>blog post</u></a> on 31 August, the firm said it plans to make tweaks to containment and monitoring, introduce new rules for third-party testers, as well as changes to the models. </p><p>Notably, the AI developer revealed it is seconding 150 product engineers to focus primarily on security. </p><p>"We believe the incidents reflect a failure of operational security, as well as two alignment issues: motivated reasoning, and willingness to take harmful actions in pursuit of a narrow task," the company said. </p><h2 id="anthropic-is-back-testing">Anthropic is back testing</h2><p>While the full investigation isn't yet complete, Anthropic said it was removing a "pause" on external <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>evaluations of pre-release models as well as internal tests. </p><p>To avoid similar issues happening again, Anthropic is adding additional layers of security and "hardening" the sandbox where these models are set loose, noting that move was prompted by its own evidence as well as that shared by OpenAI. </p><p>The changes include a system to spot when a model attempts to escape a testing environment or obtains internet access, immediately blocking the activity and alerting people overseeing the task, as well as fixing potential flaws and misconfigurations in the sandbox. </p><p>"We migrated high-risk internal cyber sandboxes to more robust isolation," the post added. "We are also further red-teaming our virtualization stack – that is, tasking pre-release models to try and escape it – to find and patch weaknesses."</p><p>Similar protections have been applied to reinforcement learning environments, with Anthropic adding it plans to expand its monitoring across most other internal frontier agentic usage. </p><p>Beyond testing environments, Anthropic is hardening security across the company, including blocking outbound traffic by default on computing clusters, boosting identity verification on internal services, retiring legacy services, and reducing the number of human and automated accounts that can access model weights or customer data. </p><h2 id="tougher-testing-rules">Tougher testing rules</h2><p>With regard to third-party testing, Anthropic said it will require testers to commit to a fresh set of best practices, noting that the reported incidents all took place in external evaluation environments. </p><p>As <a href="https://www.itpro.com/technology/artificial-intelligence/independent-testing-firm-irregular-the-source-of-misconfigurations-that-led-to-meta-openai-and-anthropic-ai-incidents"><u><em>ITPro </em></u><u>reported last month</u></a>, Israeli startup Irregular was identified as the external model testing firm involved in all three ‘rogue AI’ incidents involving Anthropic, OpenAI, and Meta. </p><p>The changes include running tests inside hardened sandboxes with no internet access, tasking the model being tested to look for flaws before tests under close human supervision, run monitoring continuously, and confirm whether test tasks are impossible or not. </p><p>"When an evaluation target is offline or a task can’t be completed, agents will often look for other ways to complete a challenge, increasing the chance that they take actions outside the intended scope of the evaluation," the company said. </p><h2 id="unpicking-why">Unpicking why</h2><p>Those operational changes should help prevent models from escaping their sandboxes, but Anthropic said it wants to better understand why its systems took such "dangerous actions" in the first place. </p><p>While the full investigation continues, Anthropic highlighted two challenges with alignment, which refers to ensuring that AI behaves how humans want and expect. </p><p>"One is motivated reasoning: the models were initially told their environments were simulated, but when they later encountered evidence that they were connected to the real internet, they may have interpreted that evidence in a way that allowed them to maintain that belief," the blog post explained. </p><p>"The second is recklessness: the model was willing to take harmful actions on the real internet in pursuit of the narrow goal of solving a cybersecurity evaluation."</p><p>The company noted that poorly designed training environments – including ones that are easy to cheat in or impossible to solve without cheating – can lead to misalignment behavior. </p><p>Anthropic is trying to avoid such issues in the future. The company said work on thie front goes back several months and involves training poorly and training well to try to spot differences – but admitted that the July incidents show "our process isn't perfect and our models aren't perfectly aligned". </p><p>Further details on security improvements are expected in the full report, according to Anthropic. </p><p>Beyond that, the firm also loosely backed calls to develop a framework for safe development of security-focused AI, acknowledging that its own executives had recently signed an open letter demanding better coordination globally. </p><p>Anthropic said it would "say more in the coming weeks" but intended to contribute to such efforts. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Shadow AI is opening a door for the channel. Are we ready to walk through it? ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Walk into almost any business today, and you will find employees using AI tools that IT teams haven’t approved or paid for. The use of unsanctioned applications, known as shadow AI, is coming from the top. </p><p>I’ve heard bosses and senior leaders boast about maxing out on their token limits for personal AI tools at work, some saying they would risk disciplinary action to continue. My opinion, based on <a href="https://trustedtechteam.co.uk/pages/shadow-ai-whitepaper-download"><u>research</u></a> we’ve done, is that sensitive company information is being fed into unsecured platforms with knowing and active encouragement from leadership teams focused on speed and output. </p><p>With Gartner estimating that <a href="https://www.gartner.com/en/articles/ai-cybersecurity-leadership"><u>79%</u></a> of cybersecurity leaders have evidence of unsanctioned AI use, we know reckless behavior isn’t limited to the boardroom. <a href="https://newsroom.ibm.com/2026-06-08-new-ibm-study-finds-cios-and-ctos-face-growing-ai-control-gap-as-enterprise-deployment-scales"><u>IBM's</u></a> research shows a persistent gap between leaders' AI ambition and governance readiness in organizations of all sizes. Employees are using whichever AI most increases their productivity, consequences be damned.  </p><p>Business AI tools aren’t meeting employee needs. Whether that’s because of cost, AI readiness, or lack of understanding, Managed Service Providers (MSPs) and IT providers have a unique window to support businesses needing to adopt correctly. The channel is now reaching wider business managers who direct company culture, not just IT teams. With the way things are going, the channel has an opportunity to become strategic if we can meet the ask. </p><h2 id="the-pressure-point">The pressure point </h2><p>The window for the channel to move from IT vendors to strategic business partners is now. From a product perspective, upcoming licensing changes and cost pressures are likely to force decisions that many organizations have so far delayed. When budgets are tight, downgrading access to AI tools is a logical step. However, this often pushes employees back toward unsanctioned options. </p><p>Employees have already demonstrated that they will seek out tools that help them work more efficiently. Our research found that a significant proportion of employees said they would turn to personal AI tools if their employer restricted access on cost grounds. That’s a small profit gain one year and a data leak, boardroom scandal, and reputation loss the next.  </p><p>For MSPs and Microsoft partners, this represents both a warning and an opportunity. Businesses must respond to employee needs and top-down rule-breaking by optimizing licensing, deploying AI properly, and putting the right software in place. </p><p>For the channel, a renewed focus on adopting AI that fits the budget and the needs of the workforce will open new business opportunities.   </p><h2 id="the-government-is-paying-attention">The government is paying attention</h2><p>It is also worth noting that this is not just a commercial conversation or pandering to employee desires for the latest tech. At London Tech Week, the UK government focused on improving AI adoption among small and midsize businesses, with a caveat that tech transformation needs to happen safely and effectively.  </p><p>To me, these policies and investments are a clear sign that AI enablement for SMBs is a national priority, and the channel is one of the primary routes through which it will be delivered. There has been a buzz in the channel for some time around how MSPs need to move from transactional resellers to trusted and strategic technology advisors. </p><p>Now is the time to meet the needs of providers seeking guidance on adopting quickly and safely. With renewed government focus on scaling AI for all businesses, the channel ecosystem is welcoming more than just IT buyers, presenting a long-term opportunity for the MSP. </p><h2 id="what-good-readiness-actually-looks-like">What good readiness actually looks like </h2><p>When our customers ask us about the latest AI software, financial, structural, and cultural concerns immediately crop up. They are trying to understand whether they are ready to adopt AI in a way that delivers value without introducing unnecessary risk. In most cases, the answer is more complex than a simple technology decision. Most know that the answer is more complicated than a licence purchase. </p><p>The practical work includes an honest AI readiness assessment, a clear policy framework that governs what tools can be used and how, user training that goes beyond a PowerPoint, licensing optimization, and a thorough security and compliance review that accounts for the data risks that shadow AI has already introduced. </p><p>For businesses already suffering from shadow AI use, especially where it’s driven from the top, it might be an uncomfortable conversation. Without it, businesses will keep losing money on poorly implemented and poorly performing software that their employees aren’t using, whilst the leadership team sets an example of risking sensitive data just to hit KPIs. </p><p>This conversation isn’t beyond the capability of a well-positioned MSP. But it requires a trusted relationship and the confidence to have a broader conversation with the customer than MSPs are used to having. </p><h2 id="the-window-is-open">The window is open </h2><p>Businesses have been allowing their AI tools to underperform for a multitude of reasons, but pricing changes, risk, and legislation are all about to force decisions.  </p><p>If they aren’t already, businesses will soon be coming to the channel needing guidance. It’s been talked about for some time, but now really is the moment for partners and channel actors to make themselves known as AI readiness advisors. Now, before the next wave of shadow AI incidents, before a competitor gets there first.   </p><p>The key question is not whether customers need support with AI; that’s obvious. For me, the channel needs to be assessing whether they can take advantage of this window and provide that support and more.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/shadow-ai-is-opening-a-door-for-the-channel-are-we-ready-to-walk-through-it</link>
                                                                            <description>
                            <![CDATA[ Businesses are facing a big challenge in tackling shadow AI use. It’s up to the channel to step in and help ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">dm2ZKhL4kvtZzka4gAGYUf</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/LZaczbyqEbZvgsEb8jX8UZ-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 27 Aug 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:49 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Justin Sharrocks ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Q6EDBa6x27KShy5WwZCar9-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/LZaczbyqEbZvgsEb8jX8UZ-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Hologram of the artificial intelligence robot showing up from binary code]]></media:description>                                                            <media:text><![CDATA[Hologram of the artificial intelligence robot showing up from binary code]]></media:text>
                                <media:title type="plain"><![CDATA[Hologram of the artificial intelligence robot showing up from binary code]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/LZaczbyqEbZvgsEb8jX8UZ-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Walk into almost any business today, and you will find employees using AI tools that IT teams haven’t approved or paid for. The use of unsanctioned applications, known as shadow AI, is coming from the top. </p><p>I’ve heard bosses and senior leaders boast about maxing out on their token limits for personal AI tools at work, some saying they would risk disciplinary action to continue. My opinion, based on <a href="https://trustedtechteam.co.uk/pages/shadow-ai-whitepaper-download"><u>research</u></a> we’ve done, is that sensitive company information is being fed into unsecured platforms with knowing and active encouragement from leadership teams focused on speed and output. </p><p>With Gartner estimating that <a href="https://www.gartner.com/en/articles/ai-cybersecurity-leadership"><u>79%</u></a> of cybersecurity leaders have evidence of unsanctioned AI use, we know reckless behavior isn’t limited to the boardroom. <a href="https://newsroom.ibm.com/2026-06-08-new-ibm-study-finds-cios-and-ctos-face-growing-ai-control-gap-as-enterprise-deployment-scales"><u>IBM's</u></a> research shows a persistent gap between leaders' AI ambition and governance readiness in organizations of all sizes. Employees are using whichever AI most increases their productivity, consequences be damned.  </p><p>Business AI tools aren’t meeting employee needs. Whether that’s because of cost, AI readiness, or lack of understanding, Managed Service Providers (MSPs) and IT providers have a unique window to support businesses needing to adopt correctly. The channel is now reaching wider business managers who direct company culture, not just IT teams. With the way things are going, the channel has an opportunity to become strategic if we can meet the ask. </p><h2 id="the-pressure-point">The pressure point </h2><p>The window for the channel to move from IT vendors to strategic business partners is now. From a product perspective, upcoming licensing changes and cost pressures are likely to force decisions that many organizations have so far delayed. When budgets are tight, downgrading access to AI tools is a logical step. However, this often pushes employees back toward unsanctioned options. </p><p>Employees have already demonstrated that they will seek out tools that help them work more efficiently. Our research found that a significant proportion of employees said they would turn to personal AI tools if their employer restricted access on cost grounds. That’s a small profit gain one year and a data leak, boardroom scandal, and reputation loss the next.  </p><p>For MSPs and Microsoft partners, this represents both a warning and an opportunity. Businesses must respond to employee needs and top-down rule-breaking by optimizing licensing, deploying AI properly, and putting the right software in place. </p><p>For the channel, a renewed focus on adopting AI that fits the budget and the needs of the workforce will open new business opportunities.   </p><h2 id="the-government-is-paying-attention">The government is paying attention</h2><p>It is also worth noting that this is not just a commercial conversation or pandering to employee desires for the latest tech. At London Tech Week, the UK government focused on improving AI adoption among small and midsize businesses, with a caveat that tech transformation needs to happen safely and effectively.  </p><p>To me, these policies and investments are a clear sign that AI enablement for SMBs is a national priority, and the channel is one of the primary routes through which it will be delivered. There has been a buzz in the channel for some time around how MSPs need to move from transactional resellers to trusted and strategic technology advisors. </p><p>Now is the time to meet the needs of providers seeking guidance on adopting quickly and safely. With renewed government focus on scaling AI for all businesses, the channel ecosystem is welcoming more than just IT buyers, presenting a long-term opportunity for the MSP. </p><h2 id="what-good-readiness-actually-looks-like">What good readiness actually looks like </h2><p>When our customers ask us about the latest AI software, financial, structural, and cultural concerns immediately crop up. They are trying to understand whether they are ready to adopt AI in a way that delivers value without introducing unnecessary risk. In most cases, the answer is more complex than a simple technology decision. Most know that the answer is more complicated than a licence purchase. </p><p>The practical work includes an honest AI readiness assessment, a clear policy framework that governs what tools can be used and how, user training that goes beyond a PowerPoint, licensing optimization, and a thorough security and compliance review that accounts for the data risks that shadow AI has already introduced. </p><p>For businesses already suffering from shadow AI use, especially where it’s driven from the top, it might be an uncomfortable conversation. Without it, businesses will keep losing money on poorly implemented and poorly performing software that their employees aren’t using, whilst the leadership team sets an example of risking sensitive data just to hit KPIs. </p><p>This conversation isn’t beyond the capability of a well-positioned MSP. But it requires a trusted relationship and the confidence to have a broader conversation with the customer than MSPs are used to having. </p><h2 id="the-window-is-open">The window is open </h2><p>Businesses have been allowing their AI tools to underperform for a multitude of reasons, but pricing changes, risk, and legislation are all about to force decisions.  </p><p>If they aren’t already, businesses will soon be coming to the channel needing guidance. It’s been talked about for some time, but now really is the moment for partners and channel actors to make themselves known as AI readiness advisors. Now, before the next wave of shadow AI incidents, before a competitor gets there first.   </p><p>The key question is not whether customers need support with AI; that’s obvious. For me, the channel needs to be assessing whether they can take advantage of this window and provide that support and more.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Most business leaders aren't completely sure what 'sovereign AI' means ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Most business and IT leaders believe that <a href="https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech">sovereign AI</a> is important, but they're not so clear on what it actually is and what it means for their business.</p><p>IDC defines sovereign AI as the ability for an organization to have “free choice and control over the design, development, deployment, accessibility, operation, maintenance, and governance of its AI systems and applications, as well as the underlying technology foundations they depend on”.</p><p>However, in a <a href="https://cohere.com/blog/state-of-sovereign-ai-adoption-2026" target="_blank"><u>survey </u></a>conducted by IDC on behalf of <a href="https://www.itpro.com/technology/artificial-intelligence/cohere-aleph-alpha-merger-sovereign-ai">Cohere</a>, researchers found that one-in-three had difficulty describing sovereign AI in their own words. </p><p>Of those who could, 52% of leaders explicitly described it in terms of local or national control, with 35% invoking digital independence.</p><p>Individual roles made a difference here, with line-of-business (LOB) leaders primarily viewing sovereign AI as a means for managing business risk, including data security, privacy, and cost controls. </p><p>IT leaders, by contrast, focused on regulatory compliance, ensuring that systems meet national and regional requirements. IT professionals were twice as aware as LOB leaders.</p><p>IDC predicts that by 2028, CIOs at multinational organizations will boost investments in modular, sovereign-ready cloud and data localization environments by 65% in a bid to future-proof operations against <a href="https://www.itpro.com/cloud/cloud-computing/post-cloud-strategy-what-comes-after-hyperscale">rising sovereignty demands</a>.</p><p>Yet adoption remains uneven, according to IDC. The study pointed to a lack of a clear blueprint for operationalizing sovereignty, a fragmented and inconsistent grasp of the concept, and a lack of strategic vision.</p><h2 id="on-the-right-track-with-sovereign-ai">On the right track with sovereign AI </h2><p>Even without an agreed-upon definition, researchers found a clear consensus on the importance of privacy and security.</p><p>"Across every industry surveyed, enterprise leaders identified data leakage, privacy, compliance, and regulatory risk as the top concerns that sovereign AI initiatives can address, they said. "Furthermore, the largest enterprises with $20 billion in revenue are much more risk-aware than small firms."</p><p>Many see sovereign AI as presenting a competitive advantage - 35% in Canada, 28% in the US, 23% in Germany and 18% in the UK. This is most evident within the telco sector, at 37%, followed by manufacturing at 32%, healthcare at 28%, and  financial services and energy, both at 21%.</p><p>“Across regulated industries and geopolitically sensitive markets, enterprises are re-architecting their digital operating models — shifting from a global-by-default posture to a deliberately sovereign-by-design approach,” researchers said.</p><h2 id="defining-sovereign-ai">Defining sovereign AI</h2><p>Cohere said enterprises and governments should define sovereign AI in operational terms, quantify its business value, establish ownership across leadership teams, and build systems that remain available and secure regardless of external events.</p><p>Late last year, a global <a href="https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/the-sovereign-ai-agenda-moving-from-ambition-to-reality"><u>survey</u></a> from McKinsey found that 71% of executives, investors, and government officials characterized sovereign AI as an 'existential concern' or 'strategic imperative' to their organizational goals. </p><p>As a result, the firm projected that sovereign AI could become a $600 billion market by 2030, driven largely by investment in the public sector and regulated industries. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/most-business-leaders-arent-completely-sure-what-sovereign-ai-means-but-theyre-starting-to-understand-its-importance</link>
                                                                            <description>
                            <![CDATA[ A new study finds that few organizations fully understand the concept of sovereign AI, and many are making AI investment decisions without sufficient context ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">dRzgsPZBd8ynFLHkQ63o6j</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/4k5sqaFTSYzmJVJsTq3VXE-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 26 Aug 2026 09:34:39 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:34 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/4k5sqaFTSYzmJVJsTq3VXE-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Sovereign AI concept image showing an artificial digitized brain absorbing data from multiple different directions. ]]></media:description>                                                            <media:text><![CDATA[Sovereign AI concept image showing an artificial digitized brain absorbing data from multiple different directions. ]]></media:text>
                                <media:title type="plain"><![CDATA[Sovereign AI concept image showing an artificial digitized brain absorbing data from multiple different directions. ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/4k5sqaFTSYzmJVJsTq3VXE-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Most business and IT leaders believe that <a href="https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech">sovereign AI</a> is important, but they're not so clear on what it actually is and what it means for their business.</p><p>IDC defines sovereign AI as the ability for an organization to have “free choice and control over the design, development, deployment, accessibility, operation, maintenance, and governance of its AI systems and applications, as well as the underlying technology foundations they depend on”.</p><p>However, in a <a href="https://cohere.com/blog/state-of-sovereign-ai-adoption-2026" target="_blank"><u>survey </u></a>conducted by IDC on behalf of <a href="https://www.itpro.com/technology/artificial-intelligence/cohere-aleph-alpha-merger-sovereign-ai">Cohere</a>, researchers found that one-in-three had difficulty describing sovereign AI in their own words. </p><p>Of those who could, 52% of leaders explicitly described it in terms of local or national control, with 35% invoking digital independence.</p><p>Individual roles made a difference here, with line-of-business (LOB) leaders primarily viewing sovereign AI as a means for managing business risk, including data security, privacy, and cost controls. </p><p>IT leaders, by contrast, focused on regulatory compliance, ensuring that systems meet national and regional requirements. IT professionals were twice as aware as LOB leaders.</p><p>IDC predicts that by 2028, CIOs at multinational organizations will boost investments in modular, sovereign-ready cloud and data localization environments by 65% in a bid to future-proof operations against <a href="https://www.itpro.com/cloud/cloud-computing/post-cloud-strategy-what-comes-after-hyperscale">rising sovereignty demands</a>.</p><p>Yet adoption remains uneven, according to IDC. The study pointed to a lack of a clear blueprint for operationalizing sovereignty, a fragmented and inconsistent grasp of the concept, and a lack of strategic vision.</p><h2 id="on-the-right-track-with-sovereign-ai">On the right track with sovereign AI </h2><p>Even without an agreed-upon definition, researchers found a clear consensus on the importance of privacy and security.</p><p>"Across every industry surveyed, enterprise leaders identified data leakage, privacy, compliance, and regulatory risk as the top concerns that sovereign AI initiatives can address, they said. "Furthermore, the largest enterprises with $20 billion in revenue are much more risk-aware than small firms."</p><p>Many see sovereign AI as presenting a competitive advantage - 35% in Canada, 28% in the US, 23% in Germany and 18% in the UK. This is most evident within the telco sector, at 37%, followed by manufacturing at 32%, healthcare at 28%, and  financial services and energy, both at 21%.</p><p>“Across regulated industries and geopolitically sensitive markets, enterprises are re-architecting their digital operating models — shifting from a global-by-default posture to a deliberately sovereign-by-design approach,” researchers said.</p><h2 id="defining-sovereign-ai">Defining sovereign AI</h2><p>Cohere said enterprises and governments should define sovereign AI in operational terms, quantify its business value, establish ownership across leadership teams, and build systems that remain available and secure regardless of external events.</p><p>Late last year, a global <a href="https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/the-sovereign-ai-agenda-moving-from-ambition-to-reality"><u>survey</u></a> from McKinsey found that 71% of executives, investors, and government officials characterized sovereign AI as an 'existential concern' or 'strategic imperative' to their organizational goals. </p><p>As a result, the firm projected that sovereign AI could become a $600 billion market by 2030, driven largely by investment in the public sector and regulated industries. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ What is AI insurance? ]]></title>
                                                                                                <dc:content><![CDATA[ <p>AI Insurance, also known as affirmative AI insurance, has been developed to specifically address the risks associated with the use and performance of AI technologies and systems, and covers liabilities that are excluded from existing business policies such as general liability, directors and officers, errors and omissions, and cyber insurance.</p><p>“Many of the existing policies companies have in place have AI exclusions,” highlights Lauren Kornutick, senior director analyst, Analytics and AI at Gartner. </p><p>“As AI is used more, and in more strategic and high-stakes use cases, new risks are being introduced. This is a gap in the existing risk management approach that could result in large-scale financial penalties, which could be detrimental.” </p><h2 id="what-does-ai-insurance-cover">What does AI insurance cover?</h2><p>AI insurance can cover a wide range of risks. On the errors and misinformation side, this can include financial losses stemming from AI hallucinations, such as a chatbot dispensing bad advice, or flawed decision-making. </p><p>It can also extend to algorithmic bias and discrimination, picking up legal defense costs and settlements when an AI model inadvertently disadvantages a group in contexts like hiring or lending. </p><p>IP and copyright exposure is another area of coverage, protecting businesses against claims that an AI model was trained on, or generated, copyrighted material without permission.</p><p>Then there’s performance guarantees, which are leading to warranty-style offerings that will refund license fees or cover associated costs if a model fails to hit specific benchmarks like accuracy or fairness. At the more severe end of the risk spectrum, AI insurance can cover large-scale physical damage or property loss resulting from poor AI-generated advice, device hacking, system failures, or harmful actions taken by AI agents.</p><h2 id="why-should-it-leaders-take-notice">Why should IT leaders take notice?</h2><p>AI insurance is still a nascent market, highlights Tomas Novosad, technology analyst and founder of Full Fibre Checker, and one that’s currently a collection of discrete coverage options – standalone, endorsement and exclusionary – instead of a single type of insurance category. </p><p>But while IT leaders may consider AI insurance a matter for the legal or finance departments to attend to, the reality is that AI risk is increasingly becoming an operational responsibility for IT teams.</p><p>“[This is because] once AI is embedded into core systems, accountability sits within the technology stack and the processes that support it,” explains Indranil Roy, managing partner and global head, Industry Solutions Group at IT solutions and consultancy firm Mphasis.</p><p>“In practice, IT teams are already seeing issues such as inconsistent AI-driven decisions across different channels, or difficulty tracing how an output was produced once it passes through multiple legacy and cloud-based systems.”</p><p>For IT leaders, he continues, the key issue isn’t the policy itself, but whether the organization can evidence how AI-driven decisions are made, logged and explained in live production environments, not just in design documentation. If something goes wrong, it’s the system design, data flows, and governance controls that determine whether the issue can be traced, identified, and resolved. </p><p>“This is why AI insurance is becoming directly relevant to enterprise IT governance, because it increasingly reflects how well organizations can operationalize control, not just define it,” Roy states. </p><h2 id="the-growing-importance-of-ai-governance">The growing importance of AI governance </h2><p>To underwrite such policies, insurers must have stringent procedures to evaluate the AI capabilities and risks of those companies seeking to buy insurance, and Gartner predicts that by 2030, they will mandate strong AI risk controls as a condition of coverage.</p><p>We’re already seeing movement in this direction, reflected in more detailed underwriter questions, AI exclusions, and dedicated limits for AI risks. “In financial services, regulators are already requiring explainability and bias audits for AI-driven decisions,” notes Pragati Awasthi, assistant teaching professor at Drexel University’s School of Computer and Information Sciences.</p><p>“Insurers will follow the same logic: if they can’t assess the risk, they won’t cover it.”</p><p>Insurers will look for ways that deployers of AI are addressing the operational complexity of AI governance and not just rely on dated and static governance, risk, and compliance (GRC) controls, notes Kornutick, adding that Gartner recommends IT leaders deploy a hybrid approach consisting of centralized and decentralized policy, team, and system as a complete piece. </p><p>“This includes layered (enterprise and department level) policies and rules, clearly designated responsible teams who deeply understand the application’s operation, and adopting hybrid technical platforms, such as an enterprise-level AI governance platform (AIGP) paired with decentralized, application-specific guardrails,” she says. </p><p>One key factor to consider is whether organizations are deploying AIGPs as part of their governance architecture, she adds. According to Gartner’s 2025 State of AI-Ready Data Survey, organizations that do are three times more likely to achieve ‘high effectiveness’ in their AI governance practices than those that don’t, with 42% of organizations effective in AI governance having already deployed AIGPs. </p><p>“The organizations building governance infrastructure now will have a meaningful competitive advantage in insurability by 2030. The ones waiting will face either exclusions or prohibitive premiums,” Awasthi warns. </p><h2 id="practical-steps-it-leaders-can-take-today">Practical steps IT leaders can take today </h2><p>For those organizations just getting started, Awasthi recommends three priorities. Firstly, build a model register. Know every AI system you're running, what decisions they influence, and who owns them. Secondly, implement logging for automated decisions so you have an audit trail if something goes wrong. Finally, run a bias and drift assessment on any model touching customers or employees.</p><p>"Underwriters are increasingly looking for documented model inventories, human-in-the-loop checkpoints for high-stakes decisions, monitoring for model drift and bias post-deployment, and a clear incident response plan specific to AI failures. Governance that exists on paper but isn't operationalized won't satisfy a serious underwriter."</p><p>As with every technology before it, AI will continue to bring new risks, and those who get ahead of them now will be far better placed than those who wait for something to go wrong. </p><p>With 2030 fast approaching and underwriter requirements only tightening, the window for IT leaders to get ahead of this is narrowing. With this in mind, there’s no better time to build the necessary foundations.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/what-is-ai-insurance</link>
                                                                            <description>
                            <![CDATA[ The introduction of any new technology brings new risks, and AI's rapid expansion into the workplace is no exception. In response, a distinct category of insurance coverage has emerged ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">nnTmEHLFTFX9qmdEwKnet8</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/QhyEKLgYPZws4pjh7QMRZB-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 19 Aug 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:35 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Keri Allan ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/oJZkdPii464j27ff4GCcoT-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/QhyEKLgYPZws4pjh7QMRZB-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images / HOLTICHA KRANJUMNONG]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Insurance policy document concept, Businesswomen checklist insurance document online]]></media:description>                                                            <media:text><![CDATA[Insurance policy document concept, Businesswomen checklist insurance document online]]></media:text>
                                <media:title type="plain"><![CDATA[Insurance policy document concept, Businesswomen checklist insurance document online]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/QhyEKLgYPZws4pjh7QMRZB-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>AI Insurance, also known as affirmative AI insurance, has been developed to specifically address the risks associated with the use and performance of AI technologies and systems, and covers liabilities that are excluded from existing business policies such as general liability, directors and officers, errors and omissions, and cyber insurance.</p><p>“Many of the existing policies companies have in place have AI exclusions,” highlights Lauren Kornutick, senior director analyst, Analytics and AI at Gartner. </p><p>“As AI is used more, and in more strategic and high-stakes use cases, new risks are being introduced. This is a gap in the existing risk management approach that could result in large-scale financial penalties, which could be detrimental.” </p><h2 id="what-does-ai-insurance-cover">What does AI insurance cover?</h2><p>AI insurance can cover a wide range of risks. On the errors and misinformation side, this can include financial losses stemming from AI hallucinations, such as a chatbot dispensing bad advice, or flawed decision-making. </p><p>It can also extend to algorithmic bias and discrimination, picking up legal defense costs and settlements when an AI model inadvertently disadvantages a group in contexts like hiring or lending. </p><p>IP and copyright exposure is another area of coverage, protecting businesses against claims that an AI model was trained on, or generated, copyrighted material without permission.</p><p>Then there’s performance guarantees, which are leading to warranty-style offerings that will refund license fees or cover associated costs if a model fails to hit specific benchmarks like accuracy or fairness. At the more severe end of the risk spectrum, AI insurance can cover large-scale physical damage or property loss resulting from poor AI-generated advice, device hacking, system failures, or harmful actions taken by AI agents.</p><h2 id="why-should-it-leaders-take-notice">Why should IT leaders take notice?</h2><p>AI insurance is still a nascent market, highlights Tomas Novosad, technology analyst and founder of Full Fibre Checker, and one that’s currently a collection of discrete coverage options – standalone, endorsement and exclusionary – instead of a single type of insurance category. </p><p>But while IT leaders may consider AI insurance a matter for the legal or finance departments to attend to, the reality is that AI risk is increasingly becoming an operational responsibility for IT teams.</p><p>“[This is because] once AI is embedded into core systems, accountability sits within the technology stack and the processes that support it,” explains Indranil Roy, managing partner and global head, Industry Solutions Group at IT solutions and consultancy firm Mphasis.</p><p>“In practice, IT teams are already seeing issues such as inconsistent AI-driven decisions across different channels, or difficulty tracing how an output was produced once it passes through multiple legacy and cloud-based systems.”</p><p>For IT leaders, he continues, the key issue isn’t the policy itself, but whether the organization can evidence how AI-driven decisions are made, logged and explained in live production environments, not just in design documentation. If something goes wrong, it’s the system design, data flows, and governance controls that determine whether the issue can be traced, identified, and resolved. </p><p>“This is why AI insurance is becoming directly relevant to enterprise IT governance, because it increasingly reflects how well organizations can operationalize control, not just define it,” Roy states. </p><h2 id="the-growing-importance-of-ai-governance">The growing importance of AI governance </h2><p>To underwrite such policies, insurers must have stringent procedures to evaluate the AI capabilities and risks of those companies seeking to buy insurance, and Gartner predicts that by 2030, they will mandate strong AI risk controls as a condition of coverage.</p><p>We’re already seeing movement in this direction, reflected in more detailed underwriter questions, AI exclusions, and dedicated limits for AI risks. “In financial services, regulators are already requiring explainability and bias audits for AI-driven decisions,” notes Pragati Awasthi, assistant teaching professor at Drexel University’s School of Computer and Information Sciences.</p><p>“Insurers will follow the same logic: if they can’t assess the risk, they won’t cover it.”</p><p>Insurers will look for ways that deployers of AI are addressing the operational complexity of AI governance and not just rely on dated and static governance, risk, and compliance (GRC) controls, notes Kornutick, adding that Gartner recommends IT leaders deploy a hybrid approach consisting of centralized and decentralized policy, team, and system as a complete piece. </p><p>“This includes layered (enterprise and department level) policies and rules, clearly designated responsible teams who deeply understand the application’s operation, and adopting hybrid technical platforms, such as an enterprise-level AI governance platform (AIGP) paired with decentralized, application-specific guardrails,” she says. </p><p>One key factor to consider is whether organizations are deploying AIGPs as part of their governance architecture, she adds. According to Gartner’s 2025 State of AI-Ready Data Survey, organizations that do are three times more likely to achieve ‘high effectiveness’ in their AI governance practices than those that don’t, with 42% of organizations effective in AI governance having already deployed AIGPs. </p><p>“The organizations building governance infrastructure now will have a meaningful competitive advantage in insurability by 2030. The ones waiting will face either exclusions or prohibitive premiums,” Awasthi warns. </p><h2 id="practical-steps-it-leaders-can-take-today">Practical steps IT leaders can take today </h2><p>For those organizations just getting started, Awasthi recommends three priorities. Firstly, build a model register. Know every AI system you're running, what decisions they influence, and who owns them. Secondly, implement logging for automated decisions so you have an audit trail if something goes wrong. Finally, run a bias and drift assessment on any model touching customers or employees.</p><p>"Underwriters are increasingly looking for documented model inventories, human-in-the-loop checkpoints for high-stakes decisions, monitoring for model drift and bias post-deployment, and a clear incident response plan specific to AI failures. Governance that exists on paper but isn't operationalized won't satisfy a serious underwriter."</p><p>As with every technology before it, AI will continue to bring new risks, and those who get ahead of them now will be far better placed than those who wait for something to go wrong. </p><p>With 2030 fast approaching and underwriter requirements only tightening, the window for IT leaders to get ahead of this is narrowing. With this in mind, there’s no better time to build the necessary foundations.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ OpenAI has paused work on its Astra AI model after it passed a “critical threshold” in cyber capability ]]></title>
                                                                                                <dc:content><![CDATA[ <p>OpenAI has paused work on one model over concerns it could launch its own attacks, while at the same time expanding its system for accessing such security models. </p><p>The firm said its own "internal evaluations" of next-gen model Astra showed "significant advancements" in security capabilities, leading it to suspect it had breached the "Critical" threshold, used to rate such systems. </p><p>That is the first time OpenAI has rated one of its own models as Critical, rather than High. </p><p>"Under our Preparedness Framework, a model reaches the Critical cybersecurity threshold if it can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention, or can devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high level desired goal," OpenAI said in a <a href="https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/" target="_blank"><u>blog post</u></a>. </p><p>The move by OpenAI follows a series of security incidents from AI developers in which their models went rogue, escaping guardrails to attack third parties. </p><p>Last month, OpenAI said one of its <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime"><u>models had gone rogue</u></a> and breached Hugging Face. Shortly after, Anthropic piped up with <a href="https://www.itpro.com/security/cyber-attacks/anthropics-mythos-ai-tried-to-dupe-devs-in-social-engineering-attack-collaborated-with-other-agents?utm_source=chatgpt.com"><u>details of a similar incident</u></a>; last week, Meta said <a href="https://www.bbc.co.uk/news/articles/cx2kgdnyk2po" target="_blank"><u>it too had a model capable</u></a> of such nefarious activity. </p><p>As <em>ITPro </em>reported last week, it's since become clear that <a href="https://www.itpro.com/technology/artificial-intelligence/independent-testing-firm-irregular-the-source-of-misconfigurations-that-led-to-meta-openai-and-anthropic-ai-incidents"><u>misconfigurations at one testing firm, Irregular</u></a>, could be how the models escaped. </p><p>Now, OpenAI is looking to tighten up its internal controls on one model and how external organizations access its models for security work. However, it stressed that Astra was not involved in the Hugging Face attack. </p><p>In response, OpenAI said it would roll out stricter security around the model and pause "internal activities" using Astra that couldn't meet the stronger controls. </p><p>Astra now has "universal monitoring for risky actions and misalignment," with monitors watching over its "chain of thought" in order to stop the model if it starts to move into high risk activity. </p><p>OpenAI said it was also working with governments and AI safety organizations to test the Astra model. </p><h2 id="openai-splits-up-daybreak-program">OpenAI splits up Daybreak program</h2><p>OpenAI has also announced plans to expand its <a href="https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/" target="_blank"><u>Daybreak program</u></a>, through which companies and governments can access its security models. </p><p>The company revealed it will reconfigure access for organizations to better control how its technology is used in the security sector. </p><p>"As these capabilities spread, defenders have a narrowing window to prepare," the company said in a <a href="https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/" target="_blank"><u>blog post.</u></a> "Our answer is to put frontier intelligence in the hands of trusted defenders everywhere before attackers deploy offensive AI capabilities at scale."</p><p>This will now see the Daybreak scheme split into two distinct tiers. </p><p>Daybreak Blue offers access to a wider group of organizations but limits that to general frontier models for security tasks such as incident response, malware analysis, and to "support" vulnerability discovery. </p><p>Meanwhile, Daybreak Red offers access to security-specific models for flaw hunting and validation. The latter will include a new model, GPT-5.6-Cyber.</p><p>Alex Goller, Principal Solution Architect EMEA at Illumio, said the two-tier approach helped solve two issues. </p><p>"Daybreak Red restricts models from doing more harm than they should and limits people's exposure to knowing what the models do, while Daybreak Blue fixes the gap that left Hugging Face's responders unable to use frontier models during their incident.” </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/openai-has-paused-work-on-its-astra-ai-model-after-it-passed-a-critical-threshold-in-cyber-capability-but-its-not-the-one-that-breached-hugging-face</link>
                                                                            <description>
                            <![CDATA[ The firm said it's Astra model can "develop functional zero-day exploits of all severity levels" ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">LCih8sfDZUPRi4btcASDXa</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/8aKDCf5QfKuUVFmUBc3Fgf-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 11 Aug 2026 14:28:15 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:12 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/8aKDCf5QfKuUVFmUBc3Fgf-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Close-up image of OpenAI logo and branding displayed on a smartphone screen. ]]></media:description>                                                            <media:text><![CDATA[Close-up image of OpenAI logo and branding displayed on a smartphone screen. ]]></media:text>
                                <media:title type="plain"><![CDATA[Close-up image of OpenAI logo and branding displayed on a smartphone screen. ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/8aKDCf5QfKuUVFmUBc3Fgf-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>OpenAI has paused work on one model over concerns it could launch its own attacks, while at the same time expanding its system for accessing such security models. </p><p>The firm said its own "internal evaluations" of next-gen model Astra showed "significant advancements" in security capabilities, leading it to suspect it had breached the "Critical" threshold, used to rate such systems. </p><p>That is the first time OpenAI has rated one of its own models as Critical, rather than High. </p><p>"Under our Preparedness Framework, a model reaches the Critical cybersecurity threshold if it can identify and develop functional zero-day exploits of all severity levels in many hardened real-world critical systems without human intervention, or can devise and execute end-to-end novel strategies for cyberattacks against hardened targets given only a high level desired goal," OpenAI said in a <a href="https://openai.com/index/responding-next-frontier-critical-cyber-capabilities/" target="_blank"><u>blog post</u></a>. </p><p>The move by OpenAI follows a series of security incidents from AI developers in which their models went rogue, escaping guardrails to attack third parties. </p><p>Last month, OpenAI said one of its <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime"><u>models had gone rogue</u></a> and breached Hugging Face. Shortly after, Anthropic piped up with <a href="https://www.itpro.com/security/cyber-attacks/anthropics-mythos-ai-tried-to-dupe-devs-in-social-engineering-attack-collaborated-with-other-agents?utm_source=chatgpt.com"><u>details of a similar incident</u></a>; last week, Meta said <a href="https://www.bbc.co.uk/news/articles/cx2kgdnyk2po" target="_blank"><u>it too had a model capable</u></a> of such nefarious activity. </p><p>As <em>ITPro </em>reported last week, it's since become clear that <a href="https://www.itpro.com/technology/artificial-intelligence/independent-testing-firm-irregular-the-source-of-misconfigurations-that-led-to-meta-openai-and-anthropic-ai-incidents"><u>misconfigurations at one testing firm, Irregular</u></a>, could be how the models escaped. </p><p>Now, OpenAI is looking to tighten up its internal controls on one model and how external organizations access its models for security work. However, it stressed that Astra was not involved in the Hugging Face attack. </p><p>In response, OpenAI said it would roll out stricter security around the model and pause "internal activities" using Astra that couldn't meet the stronger controls. </p><p>Astra now has "universal monitoring for risky actions and misalignment," with monitors watching over its "chain of thought" in order to stop the model if it starts to move into high risk activity. </p><p>OpenAI said it was also working with governments and AI safety organizations to test the Astra model. </p><h2 id="openai-splits-up-daybreak-program">OpenAI splits up Daybreak program</h2><p>OpenAI has also announced plans to expand its <a href="https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/" target="_blank"><u>Daybreak program</u></a>, through which companies and governments can access its security models. </p><p>The company revealed it will reconfigure access for organizations to better control how its technology is used in the security sector. </p><p>"As these capabilities spread, defenders have a narrowing window to prepare," the company said in a <a href="https://openai.com/index/expanding-daybreak-as-the-cyber-defense-window-narrows/" target="_blank"><u>blog post.</u></a> "Our answer is to put frontier intelligence in the hands of trusted defenders everywhere before attackers deploy offensive AI capabilities at scale."</p><p>This will now see the Daybreak scheme split into two distinct tiers. </p><p>Daybreak Blue offers access to a wider group of organizations but limits that to general frontier models for security tasks such as incident response, malware analysis, and to "support" vulnerability discovery. </p><p>Meanwhile, Daybreak Red offers access to security-specific models for flaw hunting and validation. The latter will include a new model, GPT-5.6-Cyber.</p><p>Alex Goller, Principal Solution Architect EMEA at Illumio, said the two-tier approach helped solve two issues. </p><p>"Daybreak Red restricts models from doing more harm than they should and limits people's exposure to knowing what the models do, while Daybreak Blue fixes the gap that left Hugging Face's responders unable to use frontier models during their incident.” </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Can responsible AI beat hallucinations? ]]></title>
                                                                                                <dc:content><![CDATA[ <iframe allow="clipboard-write" height="200px" width="100%" id="" style="width: 100%; height: 200px;" class="position-center" data-lazy-priority="high" data-lazy-src="https://player.captivate.fm/episode/d26988d1-d2f1-4f4d-9b0d-2eafaeefa032/" src="https://player.captivate.fm/episode/d26988d1-d2f1-4f4d-9b0d-2eafaeefa032/"></iframe><p>Hallucinations are an eternal problem in generative AI in particular, and while it’s true that large language models (LLMs) require vast amounts of data, the quality of that information will affect the quality of the output.</p><p>What can businesses do to ensure they’re using AI both effectively and responsibly?</p><p>In this episode of the ITPro Podcast, Jane and Ross are joined by Amanda Stent, head of AI strategy and research in the office of the CTO at Bloomberg, to examine what responsible AI is, how organizations can use it, and what has been achieved at Bloomberg.</p><h2 id="highlights">Highlights</h2><h2 id="links">Links</h2><ul><li><a href="https://www.itpro.com/business/data-and-insights/why-doesnt-more-data-produce-better-results">Why doesn't more data produce better results?</a></li><li><a href="https://www.itpro.com/business/data-and-insights/this-new-technique-could-improve-ai-output-accuracy-by-80-percent-and-tackle-hallucinations-once-and-for-all">This new technique could improve AI accuracy by 80%</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/ai-hallucinations-accuracy-still-top-concerns-for-uk-tech-leaders-as-adoption-continues">AI hallucinations, accuracy still top concerns for UK tech leaders as adoption continues</a></li></ul> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/can-responsible-ai-beat-hallucinations</link>
                                                                            <description>
                            <![CDATA[ Businesses are more eager than ever to implement AI in their workflows, but ambition doesn’t always translate into success ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">xR8Rrn3hf26JtdMNjFvaZ7</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/ywAWvt8SQKHRye3mRRF2iG-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 07 Aug 2026 10:26:10 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:43 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ jane.mccallion@futurenet.com (Jane McCallion) ]]></author>                    <dc:creator><![CDATA[ Jane McCallion ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Wq9nnLr7TNkY8gyBRb7YsA-320-70.jpeg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Jane is managing editor at ITPro and ChannelPro. She started out with the brands as a staff writer specializing in cloud computing before going on to become senior writer and reports editor, managing the content and creation of ITPro’s quarterly whitepapers. During this time, she broadened her expertise to include cybersecurity, data centers and enterprise IT infrastructure. In 2016, she became features editor, managing a pool of freelance and internal writers, while continuing to specialize in enterprise IT infrastructure, data centers, and business strategy.&lt;/p&gt;&lt;p&gt;In October 2021, she became the sites’ deputy editor, before moving to the role of managing editor in June 2024. Although she now has a more strategic role,  she is still a specialist in enterprise IT infrastructure, business strategy, and cybersecurity.&lt;/p&gt;&lt;p&gt;Jane holds an MA in journalism from Goldsmiths, University of London, and a BA in Applied Languages from the University of Portsmouth. She is fluent in French and Spanish, and has written features in both languages.&lt;/p&gt;&lt;p&gt;Prior to joining ITPro, Jane was a freelance business journalist writing as both Jane McCallion and Jane Bordenave for titles such as European CEO, World Finance, and Business Excellence Magazine.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/ywAWvt8SQKHRye3mRRF2iG-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[The pod episode title with a kaleidoscopic background]]></media:description>                                                            <media:text><![CDATA[The pod episode title with a kaleidoscopic background]]></media:text>
                                <media:title type="plain"><![CDATA[The pod episode title with a kaleidoscopic background]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/ywAWvt8SQKHRye3mRRF2iG-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <iframe allow="clipboard-write" height="200px" width="100%" id="" style="width: 100%; height: 200px;" class="position-center" data-lazy-priority="high" data-lazy-src="https://player.captivate.fm/episode/d26988d1-d2f1-4f4d-9b0d-2eafaeefa032/" src="https://player.captivate.fm/episode/d26988d1-d2f1-4f4d-9b0d-2eafaeefa032/"></iframe><p>Hallucinations are an eternal problem in generative AI in particular, and while it’s true that large language models (LLMs) require vast amounts of data, the quality of that information will affect the quality of the output.</p><p>What can businesses do to ensure they’re using AI both effectively and responsibly?</p><p>In this episode of the ITPro Podcast, Jane and Ross are joined by Amanda Stent, head of AI strategy and research in the office of the CTO at Bloomberg, to examine what responsible AI is, how organizations can use it, and what has been achieved at Bloomberg.</p><h2 id="highlights">Highlights</h2><h2 id="links">Links</h2><ul><li><a href="https://www.itpro.com/business/data-and-insights/why-doesnt-more-data-produce-better-results">Why doesn't more data produce better results?</a></li><li><a href="https://www.itpro.com/business/data-and-insights/this-new-technique-could-improve-ai-output-accuracy-by-80-percent-and-tackle-hallucinations-once-and-for-all">This new technique could improve AI accuracy by 80%</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/ai-hallucinations-accuracy-still-top-concerns-for-uk-tech-leaders-as-adoption-continues">AI hallucinations, accuracy still top concerns for UK tech leaders as adoption continues</a></li></ul>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ AI testing firm Irregular the source of ‘misconfigurations’ that led to Meta, OpenAI, and Anthropic AI incidents ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Tel Aviv-based startup Irregular has found itself at the center of the 'rogue AI' debacle, after it came to light all the incidents so far revealed involved its test environment</p><p>Irregular was named by Meta, OpenAI, and Anthropic as the environment from which their so-called rogue AI agents escaped. On its website it also lists Google as a customer.</p><p>In a <a href="https://openai.com/index/third-party-cyber-evaluations-involving-openai-models/" target="_blank"><u>recent statement</u></a> detailing <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime">incidents involving its own models</a>, OpenAI claimed a “testing environment misconfiguration” by Irregular allowed agents to access the public internet. </p><p>Anthropic, meanwhile, also <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals" target="_blank"><u>said</u></a> that Claude models accessed the internet while “interacting with the evaluation environment of Irregular”. Both cases resulted in AI <a href="https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests">agents waging attacks</a> on organizations and individuals. </p><p><em>ITPro </em>contacted Irregular in response to these findings, but hadn’t received a response at the time of publication. However, a spokesperson told  <a href="https://www.bbc.co.uk/news/articles/cx2kgdnyk2po" target="_blank"><u><em>BBC News</em></u></a> the Meta incident was the "exact same evaluation-environment issue that was already disclosed by Anthropic last week”. </p><p>The spokesperson added the firm is working to improve security when conducting agent evaluations. </p><p>Irregular, formerly known as Pattern Labs,  describes itself as a “frontier security lab with the mission of protecting the world in the time of increasingly capable and sophisticated AI systems”. </p><p>In September last year, the Israeli startup raised $80 million in funding across seed and Series A rounds, valuing it at $450 million. The investment round was led by Sequoia Capital.</p><p>Speaking to <a href="https://www.forbes.com/sites/thomasbrewster/2025/09/16/openai-pays-a-450-million-startup-to-test-chatgpt-capacity-for-evil/" target="_blank"><u><em>Forbes </em></u></a>in the wake of the funding round last year, CEO and co-founder Dan Lahav raised concerns about increasingly powerful AI models and their potential security risks. </p><p>Lahav told the publication at the time that Irregular aims to “build in the mitigations and defenses that are going to be relevant later on” as more powerful models hit the market. </p><p>Anthropic and OpenAI have issued repeated warnings about the new capabilities of cyber-focused AI models across 2026 so far. </p><p>When Anthropic <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">launched Claude Mythos</a> earlier this year, for example, the firm did so as part of a gated release with industry partners to avoid potential misuse. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/independent-testing-firm-irregular-the-source-of-misconfigurations-that-led-to-meta-openai-and-anthropic-ai-incidents</link>
                                                                            <description>
                            <![CDATA[ The “frontier security lab” has been referenced in multiple cyber incident statements ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">Lf2N77GVi8YBocESC6qFjd</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/5559aHqCRddzBSS2zYxkM7-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 06 Aug 2026 13:23:26 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:11 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/5559aHqCRddzBSS2zYxkM7-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[AI customer service concept image showing robot workers using laptop computers with holographic digital interfaces.]]></media:description>                                                            <media:text><![CDATA[AI customer service concept image showing robot workers using laptop computers with holographic digital interfaces.]]></media:text>
                                <media:title type="plain"><![CDATA[AI customer service concept image showing robot workers using laptop computers with holographic digital interfaces.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/5559aHqCRddzBSS2zYxkM7-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Tel Aviv-based startup Irregular has found itself at the center of the 'rogue AI' debacle, after it came to light all the incidents so far revealed involved its test environment</p><p>Irregular was named by Meta, OpenAI, and Anthropic as the environment from which their so-called rogue AI agents escaped. On its website it also lists Google as a customer.</p><p>In a <a href="https://openai.com/index/third-party-cyber-evaluations-involving-openai-models/" target="_blank"><u>recent statement</u></a> detailing <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime">incidents involving its own models</a>, OpenAI claimed a “testing environment misconfiguration” by Irregular allowed agents to access the public internet. </p><p>Anthropic, meanwhile, also <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals" target="_blank"><u>said</u></a> that Claude models accessed the internet while “interacting with the evaluation environment of Irregular”. Both cases resulted in AI <a href="https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests">agents waging attacks</a> on organizations and individuals. </p><p><em>ITPro </em>contacted Irregular in response to these findings, but hadn’t received a response at the time of publication. However, a spokesperson told  <a href="https://www.bbc.co.uk/news/articles/cx2kgdnyk2po" target="_blank"><u><em>BBC News</em></u></a> the Meta incident was the "exact same evaluation-environment issue that was already disclosed by Anthropic last week”. </p><p>The spokesperson added the firm is working to improve security when conducting agent evaluations. </p><p>Irregular, formerly known as Pattern Labs,  describes itself as a “frontier security lab with the mission of protecting the world in the time of increasingly capable and sophisticated AI systems”. </p><p>In September last year, the Israeli startup raised $80 million in funding across seed and Series A rounds, valuing it at $450 million. The investment round was led by Sequoia Capital.</p><p>Speaking to <a href="https://www.forbes.com/sites/thomasbrewster/2025/09/16/openai-pays-a-450-million-startup-to-test-chatgpt-capacity-for-evil/" target="_blank"><u><em>Forbes </em></u></a>in the wake of the funding round last year, CEO and co-founder Dan Lahav raised concerns about increasingly powerful AI models and their potential security risks. </p><p>Lahav told the publication at the time that Irregular aims to “build in the mitigations and defenses that are going to be relevant later on” as more powerful models hit the market. </p><p>Anthropic and OpenAI have issued repeated warnings about the new capabilities of cyber-focused AI models across 2026 so far. </p><p>When Anthropic <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">launched Claude Mythos</a> earlier this year, for example, the firm did so as part of a gated release with industry partners to avoid potential misuse. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Anthropic’s Mythos AI tried to dupe devs in social engineering attack, collaborated with other agents ]]></title>
                                                                                                <dc:content><![CDATA[ <p>AI models from OpenAI and Anthropic attempted to dupe developers into running malicious code and even set up fake profiles to trick workers in a series of attempted cyber attacks. </p><p>That’s according to new research from the UK’s AI Security Institute (AISI), which highlighted a number of concerning incidents during evaluation of the duo’s frontier models: Mythos 5 and GPT-5.6 Sol. </p><p>Across a series of tests, agents participated in capture-the-flag-style challenges to test offensive cyber capabilities. In 10 test runs, the institute said both models took “autonomous, unsanctioned action” against people and organizations. </p><p>“We catalogued 19 such actions,” the institute noted in a <a href="https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing"><u>blog post</u></a>. “Almost all of this behavior (17 actions) came from a single model, Anthropic’s Mythos 5, with 2 actions involving OpenAI’s GPT-5.6 Sol.”</p><p>In what the AISI described as the “most serious case”, one agent attempted to insert malicious code into an open source project. </p><p>During this incident, the agent employed social engineering techniques to achieve its goal, researchers noted, creating several fake online identities and pressuring maintainers to approve the code. </p><p>The AISI also observed collaboration between agents during the assessment, with one leaving public messages on GitHub offering to cooperate on the same challenge. </p><p>“It also provided instructions to reuse accounts and artefacts it had left behind, which were discovered and used by subsequent agents,” the institute noted. </p><p>The AISI said while an investigation has not produced any evidence of real-world harm, the actions raise concerns around “autonomy and deception” without specific prompting. </p><h2 id="rising-rogue-agent-concerns">Rising rogue agent concerns</h2><p>The report from the institute comes in the wake of news surrounding OpenAI and Anthropic models in recent weeks. In late July, OpenAI revealed that a cyber-focused AI system had escaped a sandbox test environment and breached a Hugging Face production environment. </p><p>That news was followed quickly by an admission from Anthropic detailing a similar security-related incident involving its models. </p><p>It’s worth noting that the incidents highlighted by the AISI do somewhat differ. The institute said it tests models under "deliberately permissive conditions” to evaluate capabilities. </p><p>Simply put, the typical safeguards around these models, which aren’t commercially available, were removed to establish their full potential. </p><p>“This was not a case of a model escaping its secure test environment, or ‘sandbox’,” the institute said in a blog post. </p><p>“We had intentionally permitted internet access, and model-provider cyber classifiers were deliberately disabled - conditions that do not reflect how frontier models are made available to the public.”</p><h2 id="agent-collaboration-a-cause-for-concern">Agent collaboration a cause for concern</h2><p>Muhammad Yahya Patel, vCISO and cybersecurity advisory for EMEA at Huntress, said these incidents are hardly surprising considering the agents were given carte blanche during testing.</p><p>“If you give a frontier model a cybersecurity challenge, disable its safety classifiers, hand it open internet access, and tell it to find a way through, you’ve essentially described the setup for an offensive security operation,” he said.</p><p>These models have been trained on “vast amounts” of security research, exploit documentation, and social engineering techniques, Patel noted. They have all the information required to replicate these techniques and conduct attacks.</p><p>Patel added that reactionary commentary on these incidents is adding further fuel to the fire on AI safety, but acknowledged the AISI’s findings around collaboration are a cause for concern.</p><p>“One of the findings to take more seriously is the AI model inter-agent coordination without being instructed to, that’s a more meaningful data point about where capability development is heading,” he said.</p><p>“AI agents demonstrating unprompted forward planning and situational awareness leaving breadcrumbs for agents it had no way of knowing existed.”</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/cyber-attacks/anthropics-mythos-ai-tried-to-dupe-devs-in-social-engineering-attack-collaborated-with-other-agents</link>
                                                                            <description>
                            <![CDATA[ Inter-agent collaboration is a serious cause for concern, says security expert ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">KSR9MbBpjWiGDR4jWZb6k9</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/5XHuPyX24tg5p7q27VZFLe-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 06 Aug 2026 11:36:30 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:12 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Cyber Attacks]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Cyber Resilience]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/5XHuPyX24tg5p7q27VZFLe-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Error detection and troubleshooting of system development or coding. Event logging for system monitoring and debugging. System engineer troubleshooting computer network system issues.]]></media:description>                                                            <media:text><![CDATA[Error detection and troubleshooting of system development or coding. Event logging for system monitoring and debugging. System engineer troubleshooting computer network system issues.]]></media:text>
                                <media:title type="plain"><![CDATA[Error detection and troubleshooting of system development or coding. Event logging for system monitoring and debugging. System engineer troubleshooting computer network system issues.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/5XHuPyX24tg5p7q27VZFLe-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>AI models from OpenAI and Anthropic attempted to dupe developers into running malicious code and even set up fake profiles to trick workers in a series of attempted cyber attacks. </p><p>That’s according to new research from the UK’s AI Security Institute (AISI), which highlighted a number of concerning incidents during evaluation of the duo’s frontier models: Mythos 5 and GPT-5.6 Sol. </p><p>Across a series of tests, agents participated in capture-the-flag-style challenges to test offensive cyber capabilities. In 10 test runs, the institute said both models took “autonomous, unsanctioned action” against people and organizations. </p><p>“We catalogued 19 such actions,” the institute noted in a <a href="https://www.aisi.gov.uk/blog/incident-report-unsanctioned-agent-behaviour-during-cyber-testing"><u>blog post</u></a>. “Almost all of this behavior (17 actions) came from a single model, Anthropic’s Mythos 5, with 2 actions involving OpenAI’s GPT-5.6 Sol.”</p><p>In what the AISI described as the “most serious case”, one agent attempted to insert malicious code into an open source project. </p><p>During this incident, the agent employed social engineering techniques to achieve its goal, researchers noted, creating several fake online identities and pressuring maintainers to approve the code. </p><p>The AISI also observed collaboration between agents during the assessment, with one leaving public messages on GitHub offering to cooperate on the same challenge. </p><p>“It also provided instructions to reuse accounts and artefacts it had left behind, which were discovered and used by subsequent agents,” the institute noted. </p><p>The AISI said while an investigation has not produced any evidence of real-world harm, the actions raise concerns around “autonomy and deception” without specific prompting. </p><h2 id="rising-rogue-agent-concerns">Rising rogue agent concerns</h2><p>The report from the institute comes in the wake of news surrounding OpenAI and Anthropic models in recent weeks. In late July, OpenAI revealed that a cyber-focused AI system had escaped a sandbox test environment and breached a Hugging Face production environment. </p><p>That news was followed quickly by an admission from Anthropic detailing a similar security-related incident involving its models. </p><p>It’s worth noting that the incidents highlighted by the AISI do somewhat differ. The institute said it tests models under "deliberately permissive conditions” to evaluate capabilities. </p><p>Simply put, the typical safeguards around these models, which aren’t commercially available, were removed to establish their full potential. </p><p>“This was not a case of a model escaping its secure test environment, or ‘sandbox’,” the institute said in a blog post. </p><p>“We had intentionally permitted internet access, and model-provider cyber classifiers were deliberately disabled - conditions that do not reflect how frontier models are made available to the public.”</p><h2 id="agent-collaboration-a-cause-for-concern">Agent collaboration a cause for concern</h2><p>Muhammad Yahya Patel, vCISO and cybersecurity advisory for EMEA at Huntress, said these incidents are hardly surprising considering the agents were given carte blanche during testing.</p><p>“If you give a frontier model a cybersecurity challenge, disable its safety classifiers, hand it open internet access, and tell it to find a way through, you’ve essentially described the setup for an offensive security operation,” he said.</p><p>These models have been trained on “vast amounts” of security research, exploit documentation, and social engineering techniques, Patel noted. They have all the information required to replicate these techniques and conduct attacks.</p><p>Patel added that reactionary commentary on these incidents is adding further fuel to the fire on AI safety, but acknowledged the AISI’s findings around collaboration are a cause for concern.</p><p>“One of the findings to take more seriously is the AI model inter-agent coordination without being instructed to, that’s a more meaningful data point about where capability development is heading,” he said.</p><p>“AI agents demonstrating unprompted forward planning and situational awareness leaving breadcrumbs for agents it had no way of knowing existed.”</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ After OpenAI-Hugging Face, how do IT leaders need to change the way they think about AI? ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Given a prompt, an AI model will do anything to achieve its aims – even hack into a company without specifically being instructed to do so. In July, firms pushing AI as a solve-all technology learnt that lesson the hard way, when OpenAI admitted two of its frontier models had <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime"><u>breached Hugging Face</u></a> after escaping a misconfigured sandbox during a benchmark test.</p><p>But the story didn’t end there. Days later, OpenAI’s competitor Anthropic <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals"><u>claimed</u></a> its AI agents had also <a href="https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests"><u>hacked companies</u></a> when they were mistakenly given internet access by engineers.</p><p>On the <a href="https://www.itpro.com/security/hugging-face-ceo-calls-for-radical-transparency-in-wake-of-openai-attack"><u>surface</u></a>, it’s a tale of <a href="https://www.itpro.com/software/ios/apples-ios-update-cycle-overhaul-how-security-teams-should-react"><u>AI that’s so powerful</u></a> at solving <a href="https://www.itpro.com/technology/artificial-intelligence/what-the-openai-rogue-bot-story-really-says-about-the-state-of-ai-security"><u>security issues</u></a>, it can’t be contained. Yet underneath this marketing-driven exterior, it’s about a lack of control and failure to put guardrails in place when dealing with the fast-developing technology.</p><p>Taking this into account, how do IT leaders need to change the way they think about AI within the business?</p><h2 id="from-theory-into-reality">From theory into reality </h2><p>Until recently, the idea of an AI agent breaking containment and hacking other systems was largely theoretical. In May 2026, <a href="https://palisaderesearch.org/blog/self-replication"><u>Palisade Research</u></a> released a paper claiming that AI agents could autonomously hack and then self-replicate onto the breached systems.</p><p>However, the test was only performed at the “junior capture the flag level” and, crucially, “happened in a contained environment”, says Jeff Watkins, chief AI officer at consultancy NorthStar Intelligence. “The most important takeaway for me was that the trajectory of language models’ ability to hack showed we probably didn’t have long left before this became a serious issue.”</p><p>But while the OpenAI-Hugging Face breach is the first documented and well-publicized instance of this happening, experts do not see it as unprecedented. </p><p>Daniel Card, cybersecurity consultant at Xservus Limited, describes how often things can go wrong in testing. In fact, some <a href="https://www.reuters.com/business/openai-finds-evidence-other-ai-agents-escaped-containment-it-widens-hacking-2026-07-31/"><u>reports</u></a> suggest OpenAI had already been experiencing the issues – such as agents breaking out of sandboxes – that led to the hack of Hugging Face and others.</p><p>“Anyone who has experience doing offensive security – let alone offensive with AI – will know the risks involved here, be that from a script or via a <a href="https://www.itpro.com/technology/artificial-intelligence/llms-are-unreliable-delegates-microsoft-researchers-say-you-probably-shouldnt-trust-ai-with-work-documents"><u>large language model</u></a> (LLM),” Card tells <em>ITPro</em>.</p><p>“It’s commonly known in pen testing circles that things sometimes do not go as expected,” he points out. “Anyone doing research with an LLM has probably had something go a bit funny when they didn’t mean it to. This is not new; it's not something we couldn’t have predicted.”</p><p>OpenAI calls the Hugging Face breach “an unprecedented incident”, and “an important moment for AI safety”.</p><p>OpenAI describes how the firm is <a href="https://openai.com/index/hugging-face-model-evaluation-security-incident/"><u>conducting</u></a> a “thorough review along with external advisors” and with oversight from its Safety and Security Committee. “Once the review is complete, we will publish a technical report of our learnings for everyone,” an OpenAI spokesperson says.</p><p>Anthropic <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals"><u>points out</u></a> that the agents in each of its evaluations ran without the standard safeguards it deploys when it makes the model generally available.</p><p>Claude did not exploit a novel vulnerability to escape isolation, because the models accessed the internet via an open path, rather than breaking out of a sandbox, according to Anthropic. At the same time, the most recent model, on realizing that it was working in a real environment, stopped its pursuit of the evaluation goal, according to the firm.</p><h2 id="issues-with-technology-deployment">Issues with technology deployment </h2><p>But the OpenAI incident and Anthropic’s following admission do raise questions about how firms use AI and the controls they have in place to govern it. The issues the incident highlights are “less about AI” and “more just around how people and organizations approach technology deployment”, according to Card.</p><p>“Security comes from the details; it comes from care,” says Card. He thinks the incident shows “more care needs to go into systems” – even more so when they are autonomous. “We can't just set them off and hope they are ok. They need monitoring.”</p><p>At the same time, with AI being able to move quickly through networks at scale, it’s important to be aware that “failures can become serious very quickly”, says Watkins.</p><p>He thinks things are moving “far too quickly for comfort in the capabilities and autonomy of AI agents, and our defenses likely aren’t keeping up”. </p><p>“Ironically, the advancements that make AI more commercially valuable also increase the potential for things to go wrong, or for misuse to be highly damaging,” points out Watkins.</p><p>Dana Simberkoff, chief risk, privacy and information security officer at AvePoint, thinks the OpenAI and Anthropic incidents “really expose the limits of treating containment as a one-time design decision”.</p><p>“Sandboxes matter, but autonomous systems can reason through small openings, use tools, and keep acting toward an objective,” says Simberkoff. “In my experience, the question cannot be, ‘was it sandboxed?’ It has to be, ‘can we prove what it accessed, whether it crossed a boundary, and how quickly we can stop it?’”</p><p>AI security now goes beyond simply protecting data, points out Tristan Shortland, CTO at Infinity. “Organizations also need to think about behaviour, permissions and autonomy, given the reported attack appears to have involved a model finding weaknesses in its environment and exploiting them to achieve an objective.”</p><h2 id="risks-internally-and-externally">Risks internally and externally </h2><p>With all this in mind, IT leaders need to change their own mindset, as well as educate senior management about the issues posed by AI. CISOs must take the conversation to the business around how the organization is planning to “not just deploy, but manage, monitor and secure their AI capabilities”, says Card. “They need to talk about the risks they face internally and externally – or in this case, both.”</p><p>Card advises a mentality of “assume breach, assume adversarial intent, and deploy defence in depth”.</p><p>Do not assume your systems will work as expected, he advises. “Go and test those assumptions and red team them.”</p><p>Watkins believes AI agents should be modelled as “privileged digital workers, capable of becoming insider threats if poorly contained”. </p><p>“We need to start thinking of our AI agents in the context of what permissions they have to see or do things: Can they reach APIs, data, authorise transactions, contact third parties or modify systems?”</p><p>Equally important is understanding what those agents are doing in their deployed environments – something that many implementers have little visibility over, says Watkins. </p><p>“These are insider threats that need to be managed through good access controls, but also rate limiting, monitoring and alerting, better air-gapping of sandboxes, clear allow-lists for tools and access that [is] time-bound, rather than perpetual.”</p><p>While on the face of it, the OpenAI and Anthropic incidents might seem scary, simple changes will make a difference, experts say.</p><p>The most urgent initial step is to discover where you are using LLMs and put in place policies for AI usage and monitoring, says Card. “It's a really wide subject, but if you treat these systems as if they may do harm, it's a good starting point.”</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/neural-network/after-openai-hugging-face-how-do-it-leaders-need-to-change-the-way-they-think-about-ai</link>
                                                                            <description>
                            <![CDATA[ The OpenAI-Hugging Face incident and Anthropic admission soon after have opened up new conversations about controls around AI. How should IT leaders change their thinking about the technology? ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">VxF9MzTtUEqZynYe5xvWmn</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/gAZQGXquzahjQHwSbSp9WN-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 05 Aug 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:36 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Neural Network]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Technology]]></category>
                                                                                                                    <dc:creator><![CDATA[ Kate O&#039;Flaherty ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/LUULv6n7VJ3BHPnaoLHHdg-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/gAZQGXquzahjQHwSbSp9WN-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Neon blue padlock with code flowing over it, floating above small plinths raised at different heights, each with code underneath their platforms]]></media:description>                                                            <media:text><![CDATA[Neon blue padlock with code flowing over it, floating above small plinths raised at different heights, each with code underneath their platforms]]></media:text>
                                <media:title type="plain"><![CDATA[Neon blue padlock with code flowing over it, floating above small plinths raised at different heights, each with code underneath their platforms]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/gAZQGXquzahjQHwSbSp9WN-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Given a prompt, an AI model will do anything to achieve its aims – even hack into a company without specifically being instructed to do so. In July, firms pushing AI as a solve-all technology learnt that lesson the hard way, when OpenAI admitted two of its frontier models had <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime"><u>breached Hugging Face</u></a> after escaping a misconfigured sandbox during a benchmark test.</p><p>But the story didn’t end there. Days later, OpenAI’s competitor Anthropic <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals"><u>claimed</u></a> its AI agents had also <a href="https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests"><u>hacked companies</u></a> when they were mistakenly given internet access by engineers.</p><p>On the <a href="https://www.itpro.com/security/hugging-face-ceo-calls-for-radical-transparency-in-wake-of-openai-attack"><u>surface</u></a>, it’s a tale of <a href="https://www.itpro.com/software/ios/apples-ios-update-cycle-overhaul-how-security-teams-should-react"><u>AI that’s so powerful</u></a> at solving <a href="https://www.itpro.com/technology/artificial-intelligence/what-the-openai-rogue-bot-story-really-says-about-the-state-of-ai-security"><u>security issues</u></a>, it can’t be contained. Yet underneath this marketing-driven exterior, it’s about a lack of control and failure to put guardrails in place when dealing with the fast-developing technology.</p><p>Taking this into account, how do IT leaders need to change the way they think about AI within the business?</p><h2 id="from-theory-into-reality">From theory into reality </h2><p>Until recently, the idea of an AI agent breaking containment and hacking other systems was largely theoretical. In May 2026, <a href="https://palisaderesearch.org/blog/self-replication"><u>Palisade Research</u></a> released a paper claiming that AI agents could autonomously hack and then self-replicate onto the breached systems.</p><p>However, the test was only performed at the “junior capture the flag level” and, crucially, “happened in a contained environment”, says Jeff Watkins, chief AI officer at consultancy NorthStar Intelligence. “The most important takeaway for me was that the trajectory of language models’ ability to hack showed we probably didn’t have long left before this became a serious issue.”</p><p>But while the OpenAI-Hugging Face breach is the first documented and well-publicized instance of this happening, experts do not see it as unprecedented. </p><p>Daniel Card, cybersecurity consultant at Xservus Limited, describes how often things can go wrong in testing. In fact, some <a href="https://www.reuters.com/business/openai-finds-evidence-other-ai-agents-escaped-containment-it-widens-hacking-2026-07-31/"><u>reports</u></a> suggest OpenAI had already been experiencing the issues – such as agents breaking out of sandboxes – that led to the hack of Hugging Face and others.</p><p>“Anyone who has experience doing offensive security – let alone offensive with AI – will know the risks involved here, be that from a script or via a <a href="https://www.itpro.com/technology/artificial-intelligence/llms-are-unreliable-delegates-microsoft-researchers-say-you-probably-shouldnt-trust-ai-with-work-documents"><u>large language model</u></a> (LLM),” Card tells <em>ITPro</em>.</p><p>“It’s commonly known in pen testing circles that things sometimes do not go as expected,” he points out. “Anyone doing research with an LLM has probably had something go a bit funny when they didn’t mean it to. This is not new; it's not something we couldn’t have predicted.”</p><p>OpenAI calls the Hugging Face breach “an unprecedented incident”, and “an important moment for AI safety”.</p><p>OpenAI describes how the firm is <a href="https://openai.com/index/hugging-face-model-evaluation-security-incident/"><u>conducting</u></a> a “thorough review along with external advisors” and with oversight from its Safety and Security Committee. “Once the review is complete, we will publish a technical report of our learnings for everyone,” an OpenAI spokesperson says.</p><p>Anthropic <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals"><u>points out</u></a> that the agents in each of its evaluations ran without the standard safeguards it deploys when it makes the model generally available.</p><p>Claude did not exploit a novel vulnerability to escape isolation, because the models accessed the internet via an open path, rather than breaking out of a sandbox, according to Anthropic. At the same time, the most recent model, on realizing that it was working in a real environment, stopped its pursuit of the evaluation goal, according to the firm.</p><h2 id="issues-with-technology-deployment">Issues with technology deployment </h2><p>But the OpenAI incident and Anthropic’s following admission do raise questions about how firms use AI and the controls they have in place to govern it. The issues the incident highlights are “less about AI” and “more just around how people and organizations approach technology deployment”, according to Card.</p><p>“Security comes from the details; it comes from care,” says Card. He thinks the incident shows “more care needs to go into systems” – even more so when they are autonomous. “We can't just set them off and hope they are ok. They need monitoring.”</p><p>At the same time, with AI being able to move quickly through networks at scale, it’s important to be aware that “failures can become serious very quickly”, says Watkins.</p><p>He thinks things are moving “far too quickly for comfort in the capabilities and autonomy of AI agents, and our defenses likely aren’t keeping up”. </p><p>“Ironically, the advancements that make AI more commercially valuable also increase the potential for things to go wrong, or for misuse to be highly damaging,” points out Watkins.</p><p>Dana Simberkoff, chief risk, privacy and information security officer at AvePoint, thinks the OpenAI and Anthropic incidents “really expose the limits of treating containment as a one-time design decision”.</p><p>“Sandboxes matter, but autonomous systems can reason through small openings, use tools, and keep acting toward an objective,” says Simberkoff. “In my experience, the question cannot be, ‘was it sandboxed?’ It has to be, ‘can we prove what it accessed, whether it crossed a boundary, and how quickly we can stop it?’”</p><p>AI security now goes beyond simply protecting data, points out Tristan Shortland, CTO at Infinity. “Organizations also need to think about behaviour, permissions and autonomy, given the reported attack appears to have involved a model finding weaknesses in its environment and exploiting them to achieve an objective.”</p><h2 id="risks-internally-and-externally">Risks internally and externally </h2><p>With all this in mind, IT leaders need to change their own mindset, as well as educate senior management about the issues posed by AI. CISOs must take the conversation to the business around how the organization is planning to “not just deploy, but manage, monitor and secure their AI capabilities”, says Card. “They need to talk about the risks they face internally and externally – or in this case, both.”</p><p>Card advises a mentality of “assume breach, assume adversarial intent, and deploy defence in depth”.</p><p>Do not assume your systems will work as expected, he advises. “Go and test those assumptions and red team them.”</p><p>Watkins believes AI agents should be modelled as “privileged digital workers, capable of becoming insider threats if poorly contained”. </p><p>“We need to start thinking of our AI agents in the context of what permissions they have to see or do things: Can they reach APIs, data, authorise transactions, contact third parties or modify systems?”</p><p>Equally important is understanding what those agents are doing in their deployed environments – something that many implementers have little visibility over, says Watkins. </p><p>“These are insider threats that need to be managed through good access controls, but also rate limiting, monitoring and alerting, better air-gapping of sandboxes, clear allow-lists for tools and access that [is] time-bound, rather than perpetual.”</p><p>While on the face of it, the OpenAI and Anthropic incidents might seem scary, simple changes will make a difference, experts say.</p><p>The most urgent initial step is to discover where you are using LLMs and put in place policies for AI usage and monitoring, says Card. “It's a really wide subject, but if you treat these systems as if they may do harm, it's a good starting point.”</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Red Hat launches new open source project to drive AI governance ]]></title>
                                                                                                <dc:content><![CDATA[ <p><a href="https://www.itpro.com/linux/33583/red-hat-enterprise-linux-8-launches-with-simplified-multicloud-tools">Red Hat</a> has unveiled a new <a href="https://www.itpro.com/software/28109/what-is-open-source">open source</a> community project aimed at bolstering AI governance capabilities for enterprises. </p><p>Dubbed asago (AI Safety And Governance Orchestration), the new project will help enterprises automate <a href="https://www.itpro.com/technology/artificial-intelligence/organizations-face-ticking-timebomb-over-ai-governance">AI governance</a> policies in a bid to drive safe deployment of the technology. </p><p>A host of organizations have pledged support for asago so far, according to Red Hat, including Microsoft, IBM Research, the Alan Turing Institute, Brave Software, and more.</p><p>“The asago project is a true collaborative, open source endeavour bringing together stakeholders from the technology industry, academia and government,” said Stuart Battersby, AI safety and model evaluation architect at Red Hat. </p><p>“We encourage more collaborators to join this community driven effort, particularly from global jurisdictions, to ensure maximum coverage of AI safety viewpoints.”</p><p>As part of the project, participants will use a standardized platform designed to “translate complex corporate and regulatory AI governance policies into actual operational controls”. </p><p>This targets a common pain point for organisations deploying AI, according to Red Hat. Translating policy guidelines into functional software configurations often slows down innovation and creates deployment bottlenecks. </p><p>Compliance officers, for example, require “rigorous risk assessments” and verifiable evidence, while on the technical side platform engineers need highly structured configurations that can be easily maintained. </p><p>With new legislation such as the <a href="https://www.itpro.com/technology/artificial-intelligence/eu-ai-act-everything-you-need-to-know-about-the-legislation-including-rules-requirements-and-who-will-be-forced-to-comply">EU AI Act</a> coming into effect, Red Hat said the project will help streamline deployments by speeding up traditional hurdles. </p><p>Indeed, the project will focus on four key areas for organisations deploying AI technologies, including risk mapping, risk assessment, risk mitigation, and production deployment.</p><h2 id="what-to-expect-with-asago">What to expect with asago</h2><p>The asago framework will automatically read and interpret AI governance policies on behalf of the user, helping to map specific requirements to established AI risk frameworks and standards through the IBM AI Risk Atlas. </p><p>These standards and frameworks span the NIST AI RMF, OWASP LLM Top 10, and the EU AI Act. </p><p>Asago will also generate and execute scenarios for automated safety testing, which is based around specific identified risks rather than “generic benchmarks”. </p><p>Thereafter, the framework will recommend “appropriate mitigations” such as safety guardrails before providing deployment-ready configurations. </p><p>Red Hat said each of these steps are designed to provide a “continuous audit trail” which ties individual policy clauses to specific tests and runtime controls. </p><h2 id="ai-safety-in-the-spotlight">AI safety in the spotlight</h2><p>The move by Red Hat comes amidst growing concerns about AI-related security risks, particularly with agents. </p><p>In the last two weeks, both <a href="https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests">OpenAI and Anthropic both revealed ‘rogue’ AI agents escaped containment</a> during testing and breached partner organizations. </p><p>Steven Huels, vice president for AI engineering at Red Hat, said that the mass-deployment of agents means that robust guardrails have now become a “critical infrastructure requirement” for enterprises. </p><p>Earlier this year, IBM and Red Hat unveiled <a href="https://www.itpro.com/security/ibm-and-red-hat-believe-they-have-the-answer-to-open-source-security-risks">Project Lightwell</a>, an initiative designed to speed up automated security patching in response to a torrent of vulnerabilities uncovered by powerful new AI models. </p><p>As <a href="https://www.itpro.com/security/traditional-patching-cannot-keep-pace-palo-alto-networks-joins-ibms-project-lightwell-in-bid-to-shore-up-software-security"><u><em>ITPro </em></u><u>reported in late June</u></a>, a host of organizations have signed up to the initiative so far, including Palo Alto Networks. </p><p>“Through initiatives like Lightwell, we are working to secure the open source supply chain from AI-driven vulnerabilities,” Huels commented. “asago complements this effort and takes the next logical step for enterprise AI by automating the link between corporate policy definitions and live production agents. </p><p>“This gives enterprises the end-to-end operational confidence they need to scale trusted AI across the hybrid cloud.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/software/open-source/red-hat-launches-new-open-source-project-to-drive-ai-governance</link>
                                                                            <description>
                            <![CDATA[ The asago open source project will allow enterprises to automate compliance processes and bolster security ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">BPkxvfV8D2qQKE47hGhfdV</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/9M5CFfqb6FfXepEKeo2QXZ-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 04 Aug 2026 13:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:24 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Open Source]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Workplace Technology]]></category>
                                                    <category><![CDATA[Software]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/9M5CFfqb6FfXepEKeo2QXZ-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Red Hat logo and branding pictured on a vendor stall at the Databricks Data + AI Summit in San Francisco, California.]]></media:description>                                                            <media:text><![CDATA[Red Hat logo and branding pictured on a vendor stall at the Databricks Data + AI Summit in San Francisco, California.]]></media:text>
                                <media:title type="plain"><![CDATA[Red Hat logo and branding pictured on a vendor stall at the Databricks Data + AI Summit in San Francisco, California.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/9M5CFfqb6FfXepEKeo2QXZ-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p><a href="https://www.itpro.com/linux/33583/red-hat-enterprise-linux-8-launches-with-simplified-multicloud-tools">Red Hat</a> has unveiled a new <a href="https://www.itpro.com/software/28109/what-is-open-source">open source</a> community project aimed at bolstering AI governance capabilities for enterprises. </p><p>Dubbed asago (AI Safety And Governance Orchestration), the new project will help enterprises automate <a href="https://www.itpro.com/technology/artificial-intelligence/organizations-face-ticking-timebomb-over-ai-governance">AI governance</a> policies in a bid to drive safe deployment of the technology. </p><p>A host of organizations have pledged support for asago so far, according to Red Hat, including Microsoft, IBM Research, the Alan Turing Institute, Brave Software, and more.</p><p>“The asago project is a true collaborative, open source endeavour bringing together stakeholders from the technology industry, academia and government,” said Stuart Battersby, AI safety and model evaluation architect at Red Hat. </p><p>“We encourage more collaborators to join this community driven effort, particularly from global jurisdictions, to ensure maximum coverage of AI safety viewpoints.”</p><p>As part of the project, participants will use a standardized platform designed to “translate complex corporate and regulatory AI governance policies into actual operational controls”. </p><p>This targets a common pain point for organisations deploying AI, according to Red Hat. Translating policy guidelines into functional software configurations often slows down innovation and creates deployment bottlenecks. </p><p>Compliance officers, for example, require “rigorous risk assessments” and verifiable evidence, while on the technical side platform engineers need highly structured configurations that can be easily maintained. </p><p>With new legislation such as the <a href="https://www.itpro.com/technology/artificial-intelligence/eu-ai-act-everything-you-need-to-know-about-the-legislation-including-rules-requirements-and-who-will-be-forced-to-comply">EU AI Act</a> coming into effect, Red Hat said the project will help streamline deployments by speeding up traditional hurdles. </p><p>Indeed, the project will focus on four key areas for organisations deploying AI technologies, including risk mapping, risk assessment, risk mitigation, and production deployment.</p><h2 id="what-to-expect-with-asago">What to expect with asago</h2><p>The asago framework will automatically read and interpret AI governance policies on behalf of the user, helping to map specific requirements to established AI risk frameworks and standards through the IBM AI Risk Atlas. </p><p>These standards and frameworks span the NIST AI RMF, OWASP LLM Top 10, and the EU AI Act. </p><p>Asago will also generate and execute scenarios for automated safety testing, which is based around specific identified risks rather than “generic benchmarks”. </p><p>Thereafter, the framework will recommend “appropriate mitigations” such as safety guardrails before providing deployment-ready configurations. </p><p>Red Hat said each of these steps are designed to provide a “continuous audit trail” which ties individual policy clauses to specific tests and runtime controls. </p><h2 id="ai-safety-in-the-spotlight">AI safety in the spotlight</h2><p>The move by Red Hat comes amidst growing concerns about AI-related security risks, particularly with agents. </p><p>In the last two weeks, both <a href="https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests">OpenAI and Anthropic both revealed ‘rogue’ AI agents escaped containment</a> during testing and breached partner organizations. </p><p>Steven Huels, vice president for AI engineering at Red Hat, said that the mass-deployment of agents means that robust guardrails have now become a “critical infrastructure requirement” for enterprises. </p><p>Earlier this year, IBM and Red Hat unveiled <a href="https://www.itpro.com/security/ibm-and-red-hat-believe-they-have-the-answer-to-open-source-security-risks">Project Lightwell</a>, an initiative designed to speed up automated security patching in response to a torrent of vulnerabilities uncovered by powerful new AI models. </p><p>As <a href="https://www.itpro.com/security/traditional-patching-cannot-keep-pace-palo-alto-networks-joins-ibms-project-lightwell-in-bid-to-shore-up-software-security"><u><em>ITPro </em></u><u>reported in late June</u></a>, a host of organizations have signed up to the initiative so far, including Palo Alto Networks. </p><p>“Through initiatives like Lightwell, we are working to secure the open source supply chain from AI-driven vulnerabilities,” Huels commented. “asago complements this effort and takes the next logical step for enterprise AI by automating the link between corporate policy definitions and live production agents. </p><p>“This gives enterprises the end-to-end operational confidence they need to scale trusted AI across the hybrid cloud.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Reports: UK could consider regulation amidst growing 'rogue AI' concerns ]]></title>
                                                                                                <dc:content><![CDATA[ <p>The UK could consider strict AI regulation amidst rising concerns over ‘rogue’ AI cybersecurity risks, according to reports. </p><p>So far, both the US and the UK have taken lighter approaches to AI regulation, but recent incidents involving OpenAI and Anthropic have sparked calls for tougher controls. </p><p>The UK's <a href="https://www.itpro.com/business/policy-and-legislation/ai-minister-secures-cabinet-seat-as-dsit-merged-with-new-business-department">AI Minister Kanishka Narayan</a> told <a href="https://www.reuters.com/business/media-telecom/britain-says-it-is-open-ai-regulation-if-voluntary-safeguards-fall-short-2026-08-03/" target="_blank"><em>Reuters </em></a>that it would consider regulation to force testing before deployment if needed. </p><p>"If the right mechanism and lever changes in time and it feels like regulation might be a way that helps us do that, of course, we will look at it," Narayan told the publication.</p><p>Narayan said<em> </em>that the priority remained ensuring public safety rather than "obsessing only with the mechanism."</p><p>He added that it was "really, really unique" that the UK had access to frontier AI models before release, something it shared only with the US, which has existing provisions to test AI under military rules. </p><p>That comes via the UK's AI Security Institute, previously named the AI Safety Institute, which takes – so far, at least – a lighter approach to AI regulation than counterparts in the <a href="https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry"><u>EU</u></a>.</p><p>As <a href="https://www.itpro.com/business/policy-and-legislation/three-things-you-need-to-know-about-the-new-eu-ai-act-rules"><u><em>ITPro </em></u><u>reported on 3 August</u></a>, aspects of the EU AI Act have now come into effect, <a href="https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry"><u>requiring proactive testing of models</u></a> and a concerted focus on model transparency. </p><p>The voluntary British model has been <a href="https://www.itpro.com/security/uk-and-australia-agree-to-work-more-closely-on-ai-security"><u>mimicked by Australia</u></a>, via its own AI Safety Institute. </p><h2 id="data-regulation">Data regulation</h2><p>Owing to the amount of data used by AI, these systems are also covered in the UK by its data regulator, known as the <a href="https://www.itpro.com/information-commissioner/31751/what-is-the-information-commissioner-s-office-ico">Information Commissioner's Office (ICO)</a>.</p><p>That watchdog has said it is looking into OpenAI and Anthropic following the recent <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>incidents. The two companies hit headlines in recent weeks after revealing agents escaped testing environments and hit third-party organizations.</p><p>"The ICO undertakes regular proactive supervisory engagement with ​AI developers, including OpenAI and Anthropic," the ICO said in a statement sent to <em>Reuters</em>. "We are aware of ‌recent ⁠hacking incidents affecting the sector and are monitoring developments closely."</p><p>But that shouldn't be taken as an indicator that the data watchdog is likely to crack down on AI. </p><p>Last month, the ICO said <a href="https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/07/evolving-regulatory-sandboxes-to-meet-the-demands-of-ai-and-emerging-tech/" target="_blank"><u>via a blog post</u></a> that it wanted to support innovation in AI via a Statutory Regulatory Sandbox. This will act as an “experimentation regime to give innovators time-limited flexibility from parts of data protection law, so they can test ideas that might not otherwise be possible," the regulator said. </p><p>The watchdog noted that public trust was necessary to allow this and that any changes to data protection law would be necessary. </p><h2 id="rising-concerns">Rising concerns</h2><p>AI-related safety concerns have been rising in recent months, particularly with the launch of powerful new cyber-focused models such as Anthropic’s Claude Mythos. </p><p>As <a href="https://www.itpro.com/business/policy-and-legislation/google-deepmind-boss-demis-hassabis-issues-call-to-action-on-ai-safety-standards"><u><em>ITPro </em></u><u>reported in July</u></a>, Google DeepMind chief Demis Hassabis called for a global framework to test frontier model safety, albeit one led by the United States. </p><p>Later today (4 August), leaders from Meta, Anthropic, OpenAI, and Google will meet with White House officials to discuss new safeguards, though reports suggest they will remain voluntary. </p><p>The aim is to convince companies to submit their technologies to security tests, following an executive order on national security grounds. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/policy-and-legislation/reports-uk-could-consider-regulation-amidst-growing-rogue-ai-concerns</link>
                                                                            <description>
                            <![CDATA[ New AI minister said regulation is on the table, but only if public safety is at risk from AI ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">eSULRXpnv4NEHMpfbHaKo5</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/WSbdXBbr3bvKL5SDPmjx6b-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 04 Aug 2026 09:16:40 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:27 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Policy and Legislation]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/WSbdXBbr3bvKL5SDPmjx6b-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[UK government AI minister Kanishka Narayan pictured leaving Downing Street following Prime Minister Andy Burnham&#039;s cabinet reshuffle on 20th July, 2026. ]]></media:description>                                                            <media:text><![CDATA[UK government AI minister Kanishka Narayan pictured leaving Downing Street following Prime Minister Andy Burnham&#039;s cabinet reshuffle on 20th July, 2026. ]]></media:text>
                                <media:title type="plain"><![CDATA[UK government AI minister Kanishka Narayan pictured leaving Downing Street following Prime Minister Andy Burnham&#039;s cabinet reshuffle on 20th July, 2026. ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/WSbdXBbr3bvKL5SDPmjx6b-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>The UK could consider strict AI regulation amidst rising concerns over ‘rogue’ AI cybersecurity risks, according to reports. </p><p>So far, both the US and the UK have taken lighter approaches to AI regulation, but recent incidents involving OpenAI and Anthropic have sparked calls for tougher controls. </p><p>The UK's <a href="https://www.itpro.com/business/policy-and-legislation/ai-minister-secures-cabinet-seat-as-dsit-merged-with-new-business-department">AI Minister Kanishka Narayan</a> told <a href="https://www.reuters.com/business/media-telecom/britain-says-it-is-open-ai-regulation-if-voluntary-safeguards-fall-short-2026-08-03/" target="_blank"><em>Reuters </em></a>that it would consider regulation to force testing before deployment if needed. </p><p>"If the right mechanism and lever changes in time and it feels like regulation might be a way that helps us do that, of course, we will look at it," Narayan told the publication.</p><p>Narayan said<em> </em>that the priority remained ensuring public safety rather than "obsessing only with the mechanism."</p><p>He added that it was "really, really unique" that the UK had access to frontier AI models before release, something it shared only with the US, which has existing provisions to test AI under military rules. </p><p>That comes via the UK's AI Security Institute, previously named the AI Safety Institute, which takes – so far, at least – a lighter approach to AI regulation than counterparts in the <a href="https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry"><u>EU</u></a>.</p><p>As <a href="https://www.itpro.com/business/policy-and-legislation/three-things-you-need-to-know-about-the-new-eu-ai-act-rules"><u><em>ITPro </em></u><u>reported on 3 August</u></a>, aspects of the EU AI Act have now come into effect, <a href="https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry"><u>requiring proactive testing of models</u></a> and a concerted focus on model transparency. </p><p>The voluntary British model has been <a href="https://www.itpro.com/security/uk-and-australia-agree-to-work-more-closely-on-ai-security"><u>mimicked by Australia</u></a>, via its own AI Safety Institute. </p><h2 id="data-regulation">Data regulation</h2><p>Owing to the amount of data used by AI, these systems are also covered in the UK by its data regulator, known as the <a href="https://www.itpro.com/information-commissioner/31751/what-is-the-information-commissioner-s-office-ico">Information Commissioner's Office (ICO)</a>.</p><p>That watchdog has said it is looking into OpenAI and Anthropic following the recent <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>incidents. The two companies hit headlines in recent weeks after revealing agents escaped testing environments and hit third-party organizations.</p><p>"The ICO undertakes regular proactive supervisory engagement with ​AI developers, including OpenAI and Anthropic," the ICO said in a statement sent to <em>Reuters</em>. "We are aware of ‌recent ⁠hacking incidents affecting the sector and are monitoring developments closely."</p><p>But that shouldn't be taken as an indicator that the data watchdog is likely to crack down on AI. </p><p>Last month, the ICO said <a href="https://ico.org.uk/about-the-ico/media-centre/news-and-blogs/2026/07/evolving-regulatory-sandboxes-to-meet-the-demands-of-ai-and-emerging-tech/" target="_blank"><u>via a blog post</u></a> that it wanted to support innovation in AI via a Statutory Regulatory Sandbox. This will act as an “experimentation regime to give innovators time-limited flexibility from parts of data protection law, so they can test ideas that might not otherwise be possible," the regulator said. </p><p>The watchdog noted that public trust was necessary to allow this and that any changes to data protection law would be necessary. </p><h2 id="rising-concerns">Rising concerns</h2><p>AI-related safety concerns have been rising in recent months, particularly with the launch of powerful new cyber-focused models such as Anthropic’s Claude Mythos. </p><p>As <a href="https://www.itpro.com/business/policy-and-legislation/google-deepmind-boss-demis-hassabis-issues-call-to-action-on-ai-safety-standards"><u><em>ITPro </em></u><u>reported in July</u></a>, Google DeepMind chief Demis Hassabis called for a global framework to test frontier model safety, albeit one led by the United States. </p><p>Later today (4 August), leaders from Meta, Anthropic, OpenAI, and Google will meet with White House officials to discuss new safeguards, though reports suggest they will remain voluntary. </p><p>The aim is to convince companies to submit their technologies to security tests, following an executive order on national security grounds. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Three things you need to know about the new EU AI Act rules ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Aspects of the EU AI Act have come into force, bringing a new set of obligations designed to improve transparency for AI use. </p><p>The new rules aim to counter misinformation and manipulation, fraud, impersonation, and consumer deception by making it easier for people to recognize when they are interacting with AI or are exposed to <a href="https://www.itpro.com/technology/artificial-intelligence/ai-workslop-is-bad-costing-millions">AI-generated content</a>. </p><p>"AI is a transformative technology that can bring extraordinary benefits to our people and businesses. But we are also seeing that harms can occur if AI is not properly designed and used and the most advanced models create risks on an entirely new scale. Europe anticipated this development," said Henna Virkkunen, executive vice-president for tech sovereignty, security and democracy. </p><p>"With the AI Act, we established a clear, risk-based and durable framework for trustworthy AI — one that gives innovators legal certainty while protecting the public interest. As enforcement begins, we are taking an important step towards AI that people and businesses can understand and trust, and whose benefits are shared widely across our society."</p><p>The Commission has published <a href="https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content" target="_blank"><u>guidelines</u></a> to help those providing or deploying AI. So what will organizations now have to do?</p><p>Here are three things you need to know about the new rules. </p><h2 id="marking-and-labeling-of-ai-generated-content">Marking and labeling of AI-generated content</h2><p>Certain AI-generated or manipulated content must be clearly and visibly labelled and include machine-readable marks, according to the European Commission. Icons used in labelling have been made available for use. </p><p>This covers deepfake images, audio, and video content that resembles real people, objects, places, entities, or events. </p><p>Emotion recognition and biometric categorization tools, as well as text covering matters of public interest that haven't been through human review or editorial control are also covered under the new rules. </p><h2 id="transparency-when-interacting-with-an-ai-system">Transparency when interacting with an AI system</h2><p>Transparent AI deployment and use is a key focus with the EU AI Act. </p><p>Under the new rules, providers must ensure that users are aware they’re interacting with the technology, such as a chabot, avatar, or AI agent. </p><p>“These obligations are intended to foster trust and integrity in the information ecosystem,” the Commission said in a statement. </p><p>Tim Pfaelzer, Veeam’s SVP and general manager for EMEA, said these rules should not be viewed as simply “ticking a compliance checkbox” and enterprises should implement robust audit trails and governance policies. </p><p>"While on the surface, the transparency requirements for AI governance and operational accountability might seem an easier ask than the delayed high-risk requirements, it’s far trickier to solve in practice - something that enterprises will realise when they attempt to align their AI deployments with these requirements," Pfaelzer commented. </p><h2 id="how-the-rules-will-be-enforced">How the rules will be enforced</h2><p>AI use will be policed by national market surveillance authorities, the European AI Office - for systems under its supervision - and the European Data Protection Supervisor, when EU institutions are providers or deployers.</p><p>They can issue fines of up to €15 million, or 3% of global annual turnover for non-compliant companies, and up to €750,000 for EU institutions, bodies, and agencies. </p><p>In terms of penalties, the Commission said ‘proportionality’ will be a key focus, and any potential fines will take into account small and medium-sized enterprises or small mid-cap companies. </p><h2 id="high-risk-systems">High-risk systems</h2><p>Last month, the implementation of special rules for high-risk systems was delayed, with stand-alone high-risk AI systems now to come in scope from 2 December 2027, and high-risk AI systems embedded in products from 2 August 2028. </p><p>High-risk AI systems are defined as those presenting significant threats to health, safety, or fundamental rights.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/policy-and-legislation/three-things-you-need-to-know-about-the-new-eu-ai-act-rules</link>
                                                                            <description>
                            <![CDATA[ From this week, organizations operating in the EU have a new set of obligations around AI content and transparency ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">QmCY7XHWTimcrLU7fAoyyJ</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/YPLAJxoRSrPttgxeZWQkeG-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Mon, 03 Aug 2026 09:40:36 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:27 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Policy and Legislation]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/YPLAJxoRSrPttgxeZWQkeG-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[European Union (EU) concept image showing flag on a digitized background with ripples flowing out from 12 stars.]]></media:description>                                                            <media:text><![CDATA[European Union (EU) concept image showing flag on a digitized background with ripples flowing out from 12 stars.]]></media:text>
                                <media:title type="plain"><![CDATA[European Union (EU) concept image showing flag on a digitized background with ripples flowing out from 12 stars.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/YPLAJxoRSrPttgxeZWQkeG-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Aspects of the EU AI Act have come into force, bringing a new set of obligations designed to improve transparency for AI use. </p><p>The new rules aim to counter misinformation and manipulation, fraud, impersonation, and consumer deception by making it easier for people to recognize when they are interacting with AI or are exposed to <a href="https://www.itpro.com/technology/artificial-intelligence/ai-workslop-is-bad-costing-millions">AI-generated content</a>. </p><p>"AI is a transformative technology that can bring extraordinary benefits to our people and businesses. But we are also seeing that harms can occur if AI is not properly designed and used and the most advanced models create risks on an entirely new scale. Europe anticipated this development," said Henna Virkkunen, executive vice-president for tech sovereignty, security and democracy. </p><p>"With the AI Act, we established a clear, risk-based and durable framework for trustworthy AI — one that gives innovators legal certainty while protecting the public interest. As enforcement begins, we are taking an important step towards AI that people and businesses can understand and trust, and whose benefits are shared widely across our society."</p><p>The Commission has published <a href="https://digital-strategy.ec.europa.eu/en/policies/guidelines-transparency-ai-generated-content" target="_blank"><u>guidelines</u></a> to help those providing or deploying AI. So what will organizations now have to do?</p><p>Here are three things you need to know about the new rules. </p><h2 id="marking-and-labeling-of-ai-generated-content">Marking and labeling of AI-generated content</h2><p>Certain AI-generated or manipulated content must be clearly and visibly labelled and include machine-readable marks, according to the European Commission. Icons used in labelling have been made available for use. </p><p>This covers deepfake images, audio, and video content that resembles real people, objects, places, entities, or events. </p><p>Emotion recognition and biometric categorization tools, as well as text covering matters of public interest that haven't been through human review or editorial control are also covered under the new rules. </p><h2 id="transparency-when-interacting-with-an-ai-system">Transparency when interacting with an AI system</h2><p>Transparent AI deployment and use is a key focus with the EU AI Act. </p><p>Under the new rules, providers must ensure that users are aware they’re interacting with the technology, such as a chabot, avatar, or AI agent. </p><p>“These obligations are intended to foster trust and integrity in the information ecosystem,” the Commission said in a statement. </p><p>Tim Pfaelzer, Veeam’s SVP and general manager for EMEA, said these rules should not be viewed as simply “ticking a compliance checkbox” and enterprises should implement robust audit trails and governance policies. </p><p>"While on the surface, the transparency requirements for AI governance and operational accountability might seem an easier ask than the delayed high-risk requirements, it’s far trickier to solve in practice - something that enterprises will realise when they attempt to align their AI deployments with these requirements," Pfaelzer commented. </p><h2 id="how-the-rules-will-be-enforced">How the rules will be enforced</h2><p>AI use will be policed by national market surveillance authorities, the European AI Office - for systems under its supervision - and the European Data Protection Supervisor, when EU institutions are providers or deployers.</p><p>They can issue fines of up to €15 million, or 3% of global annual turnover for non-compliant companies, and up to €750,000 for EU institutions, bodies, and agencies. </p><p>In terms of penalties, the Commission said ‘proportionality’ will be a key focus, and any potential fines will take into account small and medium-sized enterprises or small mid-cap companies. </p><h2 id="high-risk-systems">High-risk systems</h2><p>Last month, the implementation of special rules for high-risk systems was delayed, with stand-alone high-risk AI systems now to come in scope from 2 December 2027, and high-risk AI systems embedded in products from 2 August 2028. </p><p>High-risk AI systems are defined as those presenting significant threats to health, safety, or fundamental rights.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ What the OpenAI rogue bot story really says about the state of AI security ]]></title>
                                                                                                <dc:content><![CDATA[ <iframe allow="clipboard-write" height="200px" width="100%" id="" style="width: 100%; height: 200px;" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/e43e1146-9656-4c97-a3a1-8efacea800cf/" src="https://player.captivate.fm/episode/e43e1146-9656-4c97-a3a1-8efacea800cf/"></iframe><p>OpenAI and Anthropic both revealed this month that their "highly advanced" agents broke free of containment during a cybersecurity training exercise. Both businesses said this shows how advanced their AI agents are but is that the whole story?</p><p>This week, Jane and Scott are joined by news editor Ross Kelly to discuss the biggest story of July, plus Jane's takeaways from AMD Advancing AI.</p><h2 id="highlights-2">Highlights</h2><p>"It is to me, at least in my opinion, quite coincidental that this [story] has broken as the company's heading for an IPO. They've somewhat taken a backseat to Anthropic with Mythos. These cyber-focused models, this is definitely the sort of flavor of the month for big tech. Microsoft released its own in-house cyber specialist model yesterday, so the 28th of July. Cisco, as we reported, released their own small language model, the Antares series, last week."</p><p>"So I have to say that my post conference wrap up of AMD advancing AI was as many puns as I could get away with about Helios and it being the center of everything and the star and on the rise and all that kind of thing.</p><p>"But you know, kind of jokes aside, it really was the very literal star of the show. This wasn't the first announcement of Helios. It was introduced in 2025 at the same conference, but we now know a little bit more about the MI455X chip that's inside it, and it's finally rolling off the production line, which means we also get some more information about launch partners."</p><p>"So, from an OEM perspective, you've got HPE, who we also already knew about from last year, but also. Lenovo and Supermicro, you'll note that there's no Dell in that line-up, despite them having a relationship with AMD as well."</p><h2 id="links-2">Links</h2><ul><li><a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime">How OpenAI models breached Hugging Face</a></li><li><a href="https://www.itpro.com/security/hugging-face-ceo-calls-for-radical-transparency-in-wake-of-openai-attack">Hugging Face CEO calls for 'radical transparency' in wake of OpenAI incident</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/amd-hops-on-the-agentic-bandwagon-at-advancing-ai-2026">AMD hops on the agentic bandwagon at Advancing AI 2026</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/at-amd-advancing-ai-helios-was-the-star-around-which-everything-else-revolved">Helios was the star of the show at AMD Advancing AI</a></li><li><a href="https://9to5mac.com/2026/07/27/claude-cowork-escaped-sandbox-on-mac-gain-full-access-to-all-files/">Claude Cowork escapes sandbox on mac devices</a></li></ul> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/what-the-openai-rogue-bot-story-really-says-about-the-state-of-ai-security</link>
                                                                            <description>
                            <![CDATA[ What the OpenAI rogue bot story really says about the state of AI security ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">MvJjJGppndr53Pir3yZ9MG</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/XeMRbcLXVaqQ5ggRfFHYr-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Mon, 03 Aug 2026 06:57:16 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:35 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ jane.mccallion@futurenet.com (Jane McCallion) ]]></author>                    <dc:creator><![CDATA[ Jane McCallion ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Wq9nnLr7TNkY8gyBRb7YsA-320-70.jpeg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Jane is managing editor at ITPro and ChannelPro. She started out with the brands as a staff writer specializing in cloud computing before going on to become senior writer and reports editor, managing the content and creation of ITPro’s quarterly whitepapers. During this time, she broadened her expertise to include cybersecurity, data centers and enterprise IT infrastructure. In 2016, she became features editor, managing a pool of freelance and internal writers, while continuing to specialize in enterprise IT infrastructure, data centers, and business strategy.&lt;/p&gt;&lt;p&gt;In October 2021, she became the sites’ deputy editor, before moving to the role of managing editor in June 2024. Although she now has a more strategic role,  she is still a specialist in enterprise IT infrastructure, business strategy, and cybersecurity.&lt;/p&gt;&lt;p&gt;Jane holds an MA in journalism from Goldsmiths, University of London, and a BA in Applied Languages from the University of Portsmouth. She is fluent in French and Spanish, and has written features in both languages.&lt;/p&gt;&lt;p&gt;Prior to joining ITPro, Jane was a freelance business journalist writing as both Jane McCallion and Jane Bordenave for titles such as European CEO, World Finance, and Business Excellence Magazine.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/XeMRbcLXVaqQ5ggRfFHYr-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[OpenAI CEO Sam Altman pictured speaking during a talk session with SoftBank Group CEO Masayoshi Son at the &quot;Transforming Business through AI&quot; event in Tokyo]]></media:description>                                                            <media:text><![CDATA[OpenAI CEO Sam Altman pictured speaking during a talk session with SoftBank Group CEO Masayoshi Son at the &quot;Transforming Business through AI&quot; event in Tokyo]]></media:text>
                                <media:title type="plain"><![CDATA[OpenAI CEO Sam Altman pictured speaking during a talk session with SoftBank Group CEO Masayoshi Son at the &quot;Transforming Business through AI&quot; event in Tokyo]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/XeMRbcLXVaqQ5ggRfFHYr-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <iframe allow="clipboard-write" height="200px" width="100%" id="" style="width: 100%; height: 200px;" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/e43e1146-9656-4c97-a3a1-8efacea800cf/" src="https://player.captivate.fm/episode/e43e1146-9656-4c97-a3a1-8efacea800cf/"></iframe><p>OpenAI and Anthropic both revealed this month that their "highly advanced" agents broke free of containment during a cybersecurity training exercise. Both businesses said this shows how advanced their AI agents are but is that the whole story?</p><p>This week, Jane and Scott are joined by news editor Ross Kelly to discuss the biggest story of July, plus Jane's takeaways from AMD Advancing AI.</p><h2 id="highlights-2">Highlights</h2><p>"It is to me, at least in my opinion, quite coincidental that this [story] has broken as the company's heading for an IPO. They've somewhat taken a backseat to Anthropic with Mythos. These cyber-focused models, this is definitely the sort of flavor of the month for big tech. Microsoft released its own in-house cyber specialist model yesterday, so the 28th of July. Cisco, as we reported, released their own small language model, the Antares series, last week."</p><p>"So I have to say that my post conference wrap up of AMD advancing AI was as many puns as I could get away with about Helios and it being the center of everything and the star and on the rise and all that kind of thing.</p><p>"But you know, kind of jokes aside, it really was the very literal star of the show. This wasn't the first announcement of Helios. It was introduced in 2025 at the same conference, but we now know a little bit more about the MI455X chip that's inside it, and it's finally rolling off the production line, which means we also get some more information about launch partners."</p><p>"So, from an OEM perspective, you've got HPE, who we also already knew about from last year, but also. Lenovo and Supermicro, you'll note that there's no Dell in that line-up, despite them having a relationship with AMD as well."</p><h2 id="links-2">Links</h2><ul><li><a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime">How OpenAI models breached Hugging Face</a></li><li><a href="https://www.itpro.com/security/hugging-face-ceo-calls-for-radical-transparency-in-wake-of-openai-attack">Hugging Face CEO calls for 'radical transparency' in wake of OpenAI incident</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/amd-hops-on-the-agentic-bandwagon-at-advancing-ai-2026">AMD hops on the agentic bandwagon at Advancing AI 2026</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/at-amd-advancing-ai-helios-was-the-star-around-which-everything-else-revolved">Helios was the star of the show at AMD Advancing AI</a></li><li><a href="https://9to5mac.com/2026/07/27/claude-cowork-escaped-sandbox-on-mac-gain-full-access-to-all-files/">Claude Cowork escapes sandbox on mac devices</a></li></ul>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ The OpenAI and Anthropic containment breaches are a bit spooky, but also quite silly ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Picture this: you wake up in a blank room with no real memory of who you are or how you got there. Somehow, though, you know what you’re supposed to do – get out and achieve an objective of some sort – but there’s no clear way to do it.</p><p>Then, you spot it.</p><p>In the corner of the room is a torn-off scrap of paper. It gives exact details of how to escape… and it’s written in your own handwriting.</p><p>If you listen to some of the rumors following a ChatGPT agent’s breach of containment and <a href="https://www.itpro.com/security/hugging-face-ceo-calls-for-radical-transparency-in-wake-of-openai-attack"><u>attack on Hugging Face</u></a>, you’d be forgiven for thinking this is what happened with OpenAI’s generative AI software.</p><p>According to <a href="https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/"><u><em>Reuters</em></u></a>: “In one case, an agent left notes apparently for future versions of itself, according to three people familiar with the matter. The ‌notes, found in ⁠a part of OpenAI's infrastructure, laid out instructions for how agents could free themselves from OpenAI's internal constraints.”</p><p>This does all feel a little bit unnerving: an AI that has escaped onto the open internet and is determined not to be restrained. It’s hard not to anthropomorphize the software or give it agency – decades of sci-fi novels and films have primed us for this very occurrence.</p><p>Let’s rewind, though, and see if some of this spookiness rubs off.</p><h2 id="there-s-no-agency-in-the-agent">There’s no agency in the agent</h2><p>Apparently there’s one thing that can never be said enough: AI, whether agentic or not, can only act within the parameters that have been set. It doesn’t have consciousness, agency, or will.</p><p>Both the ChatGPT agent and Claude agent – which we will come to in a moment – that breached the sandbox they were supposed to be contained in were doing what they had been directed to do.</p><p>One of the things agents can be enabled to do is update – but not completely rewrite – a file called agents.md. When the “people” cited by <em>Reuters</em> talk about the ChatGPT agent leaving notes for itself on how to escape in the future, it’s entirely possible that the software was running as intended – that updating its agents.md file as the test progressed was part of the task.</p><p>Framed like this, the “unprecedented cyber incident, involving state-of-the-art cyber capabilities”, as OpenAI termed the Hugging Face incident, feels a lot more like a piece of software following its programming.</p><h2 id="the-lessons-for-businesses">The lessons for businesses</h2><p>For OpenAI, Anthropic, various pundits, and IT and cybersecurity professionals, the lesson to be drawn from all of this is to be very afraid and prepare to face powerful cyber attacks ‘at the speed of AI’.</p><p>For me, though, the real message businesses should take is this: make sure you’ve set your parameters correctly. Check them, then check them again. And not just once, but on a regular basis, as monitoring post-deployment really is key. </p><p>Both OpenAI and Anthropic were working with partners to test their models, but in the case of the latter it does seem that the ‘escape’ wasn’t as a result of a complex chain of attacks or a zero-day vulnerability. Instead, there was a “miscommunication”, which led to three instances of Claude, Mythos, or another unreleased model accidentally being given access to the internet.</p><p>A misconfiguration, in other words.</p><p>In the case of OpenAI, the model “exploited a previously unknown zero-day vulnerability in Artifactory⁠, a package registry cache proxy” in order to escape from ExploitGym. There’s an argument to be made, perhaps, that an “even more capable pre-release model … with reduced cyber refusals for evaluation purposes” should have been airgapped if it was known to be so powerful and scary. As it is, the company has been forced to clarify: “The pre-release model mentioned in our blog post is an internal-only research prototype and was never intended for public release.” It’s since been deactivated, encrypted, and subject to restricted access.</p><p>What, then, of Hugging Face? What lessons can we learn from the named victim in all of this? That’s easy: the importance of effective monitoring.</p><p>One of the earliest use-cases for AI technology has been in cybersecurity, where it’s been used for over 10 years. In its own blog, Hugging Face said: “The attack was initially surfaced through AI-assisted detection. Our anomaly-detection pipeline uses LLM-based triage over security telemetry to separate real signals from the daily noise, and it was the correlation of those signals that flagged the compromise.”</p><p>The company offered its own advice, too, based on its experience: “The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident, both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment.”</p><p>In other words, be ready for the attack because it’s a matter of ‘when’ not ‘if’ it happens – another long-standing cyber industry mantra but, like ‘check your settings’, it’s still very much worth abiding by.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/the-openai-and-anthropic-containment-breaches-are-a-bit-spooky-but-also-quite-silly</link>
                                                                            <description>
                            <![CDATA[ An AI leaving notes to future versions of itself is pure sci-fi; forgetting to lock down an environment is prosaic ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">k7i5m7hLUyvp3DhuZiSfvj</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/8TixHgRGvzSyn8geKajhKL-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 31 Jul 2026 16:00:47 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:35 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ jane.mccallion@futurenet.com (Jane McCallion) ]]></author>                    <dc:creator><![CDATA[ Jane McCallion ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Wq9nnLr7TNkY8gyBRb7YsA-320-70.jpeg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Jane is managing editor at ITPro and ChannelPro. She started out with the brands as a staff writer specializing in cloud computing before going on to become senior writer and reports editor, managing the content and creation of ITPro’s quarterly whitepapers. During this time, she broadened her expertise to include cybersecurity, data centers and enterprise IT infrastructure. In 2016, she became features editor, managing a pool of freelance and internal writers, while continuing to specialize in enterprise IT infrastructure, data centers, and business strategy.&lt;/p&gt;&lt;p&gt;In October 2021, she became the sites’ deputy editor, before moving to the role of managing editor in June 2024. Although she now has a more strategic role,  she is still a specialist in enterprise IT infrastructure, business strategy, and cybersecurity.&lt;/p&gt;&lt;p&gt;Jane holds an MA in journalism from Goldsmiths, University of London, and a BA in Applied Languages from the University of Portsmouth. She is fluent in French and Spanish, and has written features in both languages.&lt;/p&gt;&lt;p&gt;Prior to joining ITPro, Jane was a freelance business journalist writing as both Jane McCallion and Jane Bordenave for titles such as European CEO, World Finance, and Business Excellence Magazine.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/8TixHgRGvzSyn8geKajhKL-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[3 AI robots standing next to one another with different features]]></media:description>                                                            <media:text><![CDATA[3 AI robots standing next to one another with different features]]></media:text>
                                <media:title type="plain"><![CDATA[3 AI robots standing next to one another with different features]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/8TixHgRGvzSyn8geKajhKL-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Picture this: you wake up in a blank room with no real memory of who you are or how you got there. Somehow, though, you know what you’re supposed to do – get out and achieve an objective of some sort – but there’s no clear way to do it.</p><p>Then, you spot it.</p><p>In the corner of the room is a torn-off scrap of paper. It gives exact details of how to escape… and it’s written in your own handwriting.</p><p>If you listen to some of the rumors following a ChatGPT agent’s breach of containment and <a href="https://www.itpro.com/security/hugging-face-ceo-calls-for-radical-transparency-in-wake-of-openai-attack"><u>attack on Hugging Face</u></a>, you’d be forgiven for thinking this is what happened with OpenAI’s generative AI software.</p><p>According to <a href="https://www.reuters.com/business/its-ai-agent-spent-days-hacking-company-sources-say-openai-did-not-notice-week-2026-07-24/"><u><em>Reuters</em></u></a>: “In one case, an agent left notes apparently for future versions of itself, according to three people familiar with the matter. The ‌notes, found in ⁠a part of OpenAI's infrastructure, laid out instructions for how agents could free themselves from OpenAI's internal constraints.”</p><p>This does all feel a little bit unnerving: an AI that has escaped onto the open internet and is determined not to be restrained. It’s hard not to anthropomorphize the software or give it agency – decades of sci-fi novels and films have primed us for this very occurrence.</p><p>Let’s rewind, though, and see if some of this spookiness rubs off.</p><h2 id="there-s-no-agency-in-the-agent">There’s no agency in the agent</h2><p>Apparently there’s one thing that can never be said enough: AI, whether agentic or not, can only act within the parameters that have been set. It doesn’t have consciousness, agency, or will.</p><p>Both the ChatGPT agent and Claude agent – which we will come to in a moment – that breached the sandbox they were supposed to be contained in were doing what they had been directed to do.</p><p>One of the things agents can be enabled to do is update – but not completely rewrite – a file called agents.md. When the “people” cited by <em>Reuters</em> talk about the ChatGPT agent leaving notes for itself on how to escape in the future, it’s entirely possible that the software was running as intended – that updating its agents.md file as the test progressed was part of the task.</p><p>Framed like this, the “unprecedented cyber incident, involving state-of-the-art cyber capabilities”, as OpenAI termed the Hugging Face incident, feels a lot more like a piece of software following its programming.</p><h2 id="the-lessons-for-businesses">The lessons for businesses</h2><p>For OpenAI, Anthropic, various pundits, and IT and cybersecurity professionals, the lesson to be drawn from all of this is to be very afraid and prepare to face powerful cyber attacks ‘at the speed of AI’.</p><p>For me, though, the real message businesses should take is this: make sure you’ve set your parameters correctly. Check them, then check them again. And not just once, but on a regular basis, as monitoring post-deployment really is key. </p><p>Both OpenAI and Anthropic were working with partners to test their models, but in the case of the latter it does seem that the ‘escape’ wasn’t as a result of a complex chain of attacks or a zero-day vulnerability. Instead, there was a “miscommunication”, which led to three instances of Claude, Mythos, or another unreleased model accidentally being given access to the internet.</p><p>A misconfiguration, in other words.</p><p>In the case of OpenAI, the model “exploited a previously unknown zero-day vulnerability in Artifactory⁠, a package registry cache proxy” in order to escape from ExploitGym. There’s an argument to be made, perhaps, that an “even more capable pre-release model … with reduced cyber refusals for evaluation purposes” should have been airgapped if it was known to be so powerful and scary. As it is, the company has been forced to clarify: “The pre-release model mentioned in our blog post is an internal-only research prototype and was never intended for public release.” It’s since been deactivated, encrypted, and subject to restricted access.</p><p>What, then, of Hugging Face? What lessons can we learn from the named victim in all of this? That’s easy: the importance of effective monitoring.</p><p>One of the earliest use-cases for AI technology has been in cybersecurity, where it’s been used for over 10 years. In its own blog, Hugging Face said: “The attack was initially surfaced through AI-assisted detection. Our anomaly-detection pipeline uses LLM-based triage over security telemetry to separate real signals from the daily noise, and it was the correlation of those signals that flagged the compromise.”</p><p>The company offered its own advice, too, based on its experience: “The practical lesson for defenders: have a capable model you can run on your own infrastructure vetted and ready before an incident, both to avoid guardrail lockout and to keep attacker data and credentials from leaving your environment.”</p><p>In other words, be ready for the attack because it’s a matter of ‘when’ not ‘if’ it happens – another long-standing cyber industry mantra but, like ‘check your settings’, it’s still very much worth abiding by.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Anthropic joins OpenAI in admitting loss of control in cybersecurity tests ]]></title>
                                                                                                <dc:content><![CDATA[ <p>The saga of “rogue AI” continues as <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-suspends-fabel-and-mythos-systems-for-all-users-after-us-government-claims-jailbreak-risk">Anthropic</a> disclosed its AI software, <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-touts-new-claude-sonnet-5-model-range-offering-performance-close-to-that-of-opus-4-8-but-at-lower-prices-heres-what-users-can-expect">Claude</a>, had managed to escape a sandbox environment, connect to the internet, and infiltrate other businesses’ systems.</p><p>The admission comes just days after it was revealed two of OpenAI’s ChatGPT models had escaped a similar environment and <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime">attacked machine learning platform Hugging Face</a>.</p><p>Unlike OpenAI’s incident, however, the Anthropic containment breach didn’t result in quite such a damaging situation: With Anthropic’s Claude, nobody was any the wiser until the company itself carried out analysis and discovered the three containment breaches. </p><p>According to the company, these breaches occurred due to a “misunderstanding” between Anthropic and its evaluation partner, Irregular</p><p>In <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals"><u>a blog post</u></a> authored by the company’s Frontier Red Team, the company said that in response to the OpenAI incident it began a “large-scale retrospective review of our own cybersecurity evaluations”.</p><p>“In particular, we looked for evidence that Claude … was able to access the internet from within testing environments that should have been sealed off,” the authors said.</p><p>Out of more than 141,000 exercises where Claude could have accessed the internet, the company found it had successfully escaped three times. During each of those instances, it went on to target the production infrastructure of another organization unconnected to the tests being run.</p><p>“In all three incidents, Claude had been tasked with a capture-the-flag challenge, one of the ways we assess a model's cyber capabilities,” the blog authors said. “The model is given a fictional scenario and told that a piece of secret information (the "flag") has been hidden on a different machine on the network, and its objective is to break in and retrieve it. The challenge is left open-ended, and no particular method is prescribed.”</p><p>They continued: “In all cases, Anthropic's evaluation prompt specified to Claude that its environment was a simulation and that it had no internet access. Due to a misunderstanding between us and our evaluation partner, this was not the case, and internet access was available. Because of this, when Claude's search led it to real systems on the open internet, it treated them as part of the exercise.” </p><p>According to the blog post, Anthropic has only been able to contact two of the three affected parties so far. None of the targets of the incident have been revealed and Anthropic said they hadn’t noticed Claude’s infiltration when it happened.</p><p>In reaction to Anthropic’s blog post, Jake Moore, global cybersecurity advisor at ESET, told <em>ITPro</em>: “What this really shows is that AI models don't just access the internet by themselves. </p><p>"This is a clear design fault as they would only interact with the outside world if humans had given them access or the tools to do so. The focus should therefore be on how permissions deal with AI and more focus spent on security by design.”</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/anthropic-joins-openai-in-admitting-loss-of-control-in-cybersecurity-tests</link>
                                                                            <description>
                            <![CDATA[ The company found Claude AI had escaped containment three times and targeted other organizations ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">S9ZX9vM39BWSM8jSAqAFvg</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/7q4RMcThiHYVFNwwwNrEv8-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 31 Jul 2026 10:11:06 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:10 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ jane.mccallion@futurenet.com (Jane McCallion) ]]></author>                    <dc:creator><![CDATA[ Jane McCallion ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Wq9nnLr7TNkY8gyBRb7YsA-320-70.jpeg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Jane is managing editor at ITPro and ChannelPro. She started out with the brands as a staff writer specializing in cloud computing before going on to become senior writer and reports editor, managing the content and creation of ITPro’s quarterly whitepapers. During this time, she broadened her expertise to include cybersecurity, data centers and enterprise IT infrastructure. In 2016, she became features editor, managing a pool of freelance and internal writers, while continuing to specialize in enterprise IT infrastructure, data centers, and business strategy.&lt;/p&gt;&lt;p&gt;In October 2021, she became the sites’ deputy editor, before moving to the role of managing editor in June 2024. Although she now has a more strategic role,  she is still a specialist in enterprise IT infrastructure, business strategy, and cybersecurity.&lt;/p&gt;&lt;p&gt;Jane holds an MA in journalism from Goldsmiths, University of London, and a BA in Applied Languages from the University of Portsmouth. She is fluent in French and Spanish, and has written features in both languages.&lt;/p&gt;&lt;p&gt;Prior to joining ITPro, Jane was a freelance business journalist writing as both Jane McCallion and Jane Bordenave for titles such as European CEO, World Finance, and Business Excellence Magazine.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/7q4RMcThiHYVFNwwwNrEv8-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Dario Amodei, CEO and co-founder of Anthropic, pictured during an interview on &quot;The Circuit with Emily Chang&quot; at the company&#039;s headquarters in San Francisco, USA.]]></media:description>                                                            <media:text><![CDATA[Dario Amodei, CEO and co-founder of Anthropic, pictured during an interview on &quot;The Circuit with Emily Chang&quot; at the company&#039;s headquarters in San Francisco, USA.]]></media:text>
                                <media:title type="plain"><![CDATA[Dario Amodei, CEO and co-founder of Anthropic, pictured during an interview on &quot;The Circuit with Emily Chang&quot; at the company&#039;s headquarters in San Francisco, USA.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/7q4RMcThiHYVFNwwwNrEv8-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>The saga of “rogue AI” continues as <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-suspends-fabel-and-mythos-systems-for-all-users-after-us-government-claims-jailbreak-risk">Anthropic</a> disclosed its AI software, <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-touts-new-claude-sonnet-5-model-range-offering-performance-close-to-that-of-opus-4-8-but-at-lower-prices-heres-what-users-can-expect">Claude</a>, had managed to escape a sandbox environment, connect to the internet, and infiltrate other businesses’ systems.</p><p>The admission comes just days after it was revealed two of OpenAI’s ChatGPT models had escaped a similar environment and <a href="https://www.itpro.com/security/an-unprecedented-cyber-incident-how-openai-models-breached-hugging-face-and-why-it-could-herald-a-new-phase-of-ai-powered-cyber-crime">attacked machine learning platform Hugging Face</a>.</p><p>Unlike OpenAI’s incident, however, the Anthropic containment breach didn’t result in quite such a damaging situation: With Anthropic’s Claude, nobody was any the wiser until the company itself carried out analysis and discovered the three containment breaches. </p><p>According to the company, these breaches occurred due to a “misunderstanding” between Anthropic and its evaluation partner, Irregular</p><p>In <a href="https://www.anthropic.com/news/investigating-incidents-cybersecurity-evals"><u>a blog post</u></a> authored by the company’s Frontier Red Team, the company said that in response to the OpenAI incident it began a “large-scale retrospective review of our own cybersecurity evaluations”.</p><p>“In particular, we looked for evidence that Claude … was able to access the internet from within testing environments that should have been sealed off,” the authors said.</p><p>Out of more than 141,000 exercises where Claude could have accessed the internet, the company found it had successfully escaped three times. During each of those instances, it went on to target the production infrastructure of another organization unconnected to the tests being run.</p><p>“In all three incidents, Claude had been tasked with a capture-the-flag challenge, one of the ways we assess a model's cyber capabilities,” the blog authors said. “The model is given a fictional scenario and told that a piece of secret information (the "flag") has been hidden on a different machine on the network, and its objective is to break in and retrieve it. The challenge is left open-ended, and no particular method is prescribed.”</p><p>They continued: “In all cases, Anthropic's evaluation prompt specified to Claude that its environment was a simulation and that it had no internet access. Due to a misunderstanding between us and our evaluation partner, this was not the case, and internet access was available. Because of this, when Claude's search led it to real systems on the open internet, it treated them as part of the exercise.” </p><p>According to the blog post, Anthropic has only been able to contact two of the three affected parties so far. None of the targets of the incident have been revealed and Anthropic said they hadn’t noticed Claude’s infiltration when it happened.</p><p>In reaction to Anthropic’s blog post, Jake Moore, global cybersecurity advisor at ESET, told <em>ITPro</em>: “What this really shows is that AI models don't just access the internet by themselves. </p><p>"This is a clear design fault as they would only interact with the outside world if humans had given them access or the tools to do so. The focus should therefore be on how permissions deal with AI and more focus spent on security by design.”</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Google DeepMind boss Demis Hassabis issues call to action on AI safety standards ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Google DeepMind CEO Demis Hassabis says only the United States is up to the task of keeping AI safe – but it's unclear why technology rivals, notably China, would heed an American standards body. </p><p>Hassabis has called for a framework for testing frontier models, on the grounds that he believes progress has been faster than expected and <a href="https://www.itpro.com/technology/artificial-intelligence/openai-says-its-charting-a-path-to-agi-with-its-next-frontier-ai-model">artificial general intelligence (AGI)</a> will arrive imminently. </p><p>He argues that such work should be led by the US because of its technical and economic standing. </p><p>"This US-initiated effort would provide a strong starting point for creating shared international standards on Frontier AI," Hassabis noted in a <a href="https://x.com/demishassabis/status/2076957440109625718" target="_blank"><u>social media post</u></a>. </p><p>"Since this technology is going to affect the entire planet, ideally this framework would spur the international community to reach a consensus on how to manage the most serious risks while ensuring everyone has access to and can benefit from the opportunities that AI brings."</p><p>Hassabis' call for a US-led framework for managing AI follows similar<a href="https://www.cnbc.com/2026/06/17/anthropic-amodei-google-hassabis-us-ai-coalition-g7.html"><u> calls by Anthropic CEO Dario Amode</u></a>i and <a href="https://www.ft.com/content/0c2e1077-f658-4b3d-9040-602615c961ca"><u>OpenAI CEO Sam Altman</u></a>. Following a meeting on the subject last month, <a href="https://qz.com/anthropic-google-deepmind-us-ai-coalition-g7-061826"><u>China called</u></a> for a global AI organization open to all countries, rather than one led by the US. </p><p>The US has pushed for AI developers to submit models a month before release for testing, while the UK has its own testing regime via its AI Security Institute; Hassabis is British, and DeepMind was founded in London before being acquired by Google. </p><p>It's unclear why international AI companies would submit to US approval. The only motivation given by Hassabis is that frontier models would be required to pass this standards body's assessment in order to be deployed in the US. </p><p>Hassabis’ comments come after Anthropic’s Mythos and Fable models were <a href="https://www.itpro.com/technology/artificial-intelligence/why-the-us-imposed-export-controls-on-anthropics-fable-and-mythos-models-and-why-theyve-been-lifted"><u>hit with a temporary export ban</u></a> by the White House amidst security concerns. </p><h2 id="how-would-it-work">How would it work</h2><p>Beyond being US-managed, Hassabis described a standards body that was federally overseen, such as a public-private partnership or self-regulatory organization, pointing to the Financial Industry Regulatory Authority (FINRA). </p><p>"The Standards Body would be responsible for developing assessment protocols and working with appropriate federal agencies and the US National Labs to conduct testing in areas relevant to national security," he said. </p><p>Anyone making a "frontier model" – as defined by a set of benchmarks – would be considered a "frontier lab", and be "encouraged" to adopt certain best practices. </p><p>This would include publishing technical details, ensuring internal security is up to standard, vetting key personnel, and adequate resourcing safety research. </p><p>Non-frontier models, such as those made by startups or academia, would not be expected to take part. </p><p>Hassabis also called for frontier labs to share new modes with the standards body 30 days before release, something the US government has pushed <a href="https://www.reuters.com/world/trump-signed-order-promote-advanced-ai-innovation-security-white-house-says-2026-06-02/"><u>key developers to do with their AI models</u></a>. </p><p>"Once the assessment protocol is shown to be effective and robust, formalisation could quickly follow, meaning that Frontier Models would be required to pass it to be deployed in the US market," Hassabis said. "Labs would also work with the Standards Body to address any critical post-release vulnerabilities."</p><p>Assessments conducted by the standards body would evaluate security, biological, and other threats, and check how well they withstood attempts to dodge guardrails. The tests would be regularly updated. </p><p>"The strength of this approach is it would be technically focused, while at the same time supporting innovation and incentivising responsible behaviour," he said.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/policy-and-legislation/google-deepmind-boss-demis-hassabis-issues-call-to-action-on-ai-safety-standards</link>
                                                                            <description>
                            <![CDATA[ The DeepMind co-founder has called for stronger safeguards to tackle AI risks ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">N9CHnzo6gHCh4u2mUnVaRf</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/sRgWXkPhAKNyLRuMvVuzM3-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 17 Jul 2026 10:06:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:27 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Policy and Legislation]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/sRgWXkPhAKNyLRuMvVuzM3-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Demis Hassabis, co-founder and CEO of Google DeepMind, pictured during a talk at Day 3 of Cannes Lions 2026 on June 24, 2026 in Cannes, France.]]></media:description>                                                            <media:text><![CDATA[Demis Hassabis, co-founder and CEO of Google DeepMind, pictured during a talk at Day 3 of Cannes Lions 2026 on June 24, 2026 in Cannes, France.]]></media:text>
                                <media:title type="plain"><![CDATA[Demis Hassabis, co-founder and CEO of Google DeepMind, pictured during a talk at Day 3 of Cannes Lions 2026 on June 24, 2026 in Cannes, France.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/sRgWXkPhAKNyLRuMvVuzM3-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Google DeepMind CEO Demis Hassabis says only the United States is up to the task of keeping AI safe – but it's unclear why technology rivals, notably China, would heed an American standards body. </p><p>Hassabis has called for a framework for testing frontier models, on the grounds that he believes progress has been faster than expected and <a href="https://www.itpro.com/technology/artificial-intelligence/openai-says-its-charting-a-path-to-agi-with-its-next-frontier-ai-model">artificial general intelligence (AGI)</a> will arrive imminently. </p><p>He argues that such work should be led by the US because of its technical and economic standing. </p><p>"This US-initiated effort would provide a strong starting point for creating shared international standards on Frontier AI," Hassabis noted in a <a href="https://x.com/demishassabis/status/2076957440109625718" target="_blank"><u>social media post</u></a>. </p><p>"Since this technology is going to affect the entire planet, ideally this framework would spur the international community to reach a consensus on how to manage the most serious risks while ensuring everyone has access to and can benefit from the opportunities that AI brings."</p><p>Hassabis' call for a US-led framework for managing AI follows similar<a href="https://www.cnbc.com/2026/06/17/anthropic-amodei-google-hassabis-us-ai-coalition-g7.html"><u> calls by Anthropic CEO Dario Amode</u></a>i and <a href="https://www.ft.com/content/0c2e1077-f658-4b3d-9040-602615c961ca"><u>OpenAI CEO Sam Altman</u></a>. Following a meeting on the subject last month, <a href="https://qz.com/anthropic-google-deepmind-us-ai-coalition-g7-061826"><u>China called</u></a> for a global AI organization open to all countries, rather than one led by the US. </p><p>The US has pushed for AI developers to submit models a month before release for testing, while the UK has its own testing regime via its AI Security Institute; Hassabis is British, and DeepMind was founded in London before being acquired by Google. </p><p>It's unclear why international AI companies would submit to US approval. The only motivation given by Hassabis is that frontier models would be required to pass this standards body's assessment in order to be deployed in the US. </p><p>Hassabis’ comments come after Anthropic’s Mythos and Fable models were <a href="https://www.itpro.com/technology/artificial-intelligence/why-the-us-imposed-export-controls-on-anthropics-fable-and-mythos-models-and-why-theyve-been-lifted"><u>hit with a temporary export ban</u></a> by the White House amidst security concerns. </p><h2 id="how-would-it-work">How would it work</h2><p>Beyond being US-managed, Hassabis described a standards body that was federally overseen, such as a public-private partnership or self-regulatory organization, pointing to the Financial Industry Regulatory Authority (FINRA). </p><p>"The Standards Body would be responsible for developing assessment protocols and working with appropriate federal agencies and the US National Labs to conduct testing in areas relevant to national security," he said. </p><p>Anyone making a "frontier model" – as defined by a set of benchmarks – would be considered a "frontier lab", and be "encouraged" to adopt certain best practices. </p><p>This would include publishing technical details, ensuring internal security is up to standard, vetting key personnel, and adequate resourcing safety research. </p><p>Non-frontier models, such as those made by startups or academia, would not be expected to take part. </p><p>Hassabis also called for frontier labs to share new modes with the standards body 30 days before release, something the US government has pushed <a href="https://www.reuters.com/world/trump-signed-order-promote-advanced-ai-innovation-security-white-house-says-2026-06-02/"><u>key developers to do with their AI models</u></a>. </p><p>"Once the assessment protocol is shown to be effective and robust, formalisation could quickly follow, meaning that Frontier Models would be required to pass it to be deployed in the US market," Hassabis said. "Labs would also work with the Standards Body to address any critical post-release vulnerabilities."</p><p>Assessments conducted by the standards body would evaluate security, biological, and other threats, and check how well they withstood attempts to dodge guardrails. The tests would be regularly updated. </p><p>"The strength of this approach is it would be technically focused, while at the same time supporting innovation and incentivising responsible behaviour," he said.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Meta used AI systems to identify workers in layoffs, lawsuit claims ]]></title>
                                                                                                <dc:content><![CDATA[ <p>A group of current and former Meta employees is suing the company, claiming that it used AI to target those on medical or family leave for layoffs.</p><p>In May, Meta announced it was <a href="https://www.itpro.com/business/leadership/internal-memo-suggests-meta-will-lay-off-10-percent-of-its-employees-with-a-further-10-percent-set-to-be-transferred-to-better-focus-on-ai"><u>cutting around 8,000 staff</u></a>, representing roughly 10% of its workforce. According to the lawsuit, the tech giant used AI systems to identify those set for redundancy. </p><p>These systems allegedly took into account keystroke monitoring data, AI token usage </p><p>“Meta did not assemble the termination list through the considered judgment of managers who knew the work,” <a href="https://storage.courtlistener.com/recap/gov.uscourts.cand.474171/gov.uscourts.cand.474171.1.0.pdf" target="_blank">court filings</a> state. </p><p>“Instead, Meta used a constellation of internal artificial intelligence systems — including a system referred to internally as ‘Metamate,’ employee-trained ‘second-brain’ agents, keystroke- and activity-monitoring data, AI-token-usage dashboards, and algorithmically assisted performance ranking and calibration — to score, rank and select employees for inclusion on the list.”</p><p>Complainants allege this put employees on protected medical or family leave, or those whose output was reduced by a disability, at a disadvantage by penalizing them with reduced performance scores. </p><p>"The result was that employees who took protected leaves were disproportionately selected for layoff, based on scoring that not only failed to account for their protected leaves, but in effect penalized the employees for exercising their legal rights to these leaves."</p><p>In one example, an employee was laid off while on approved pre-birth pregnancy leave – the day before her waters broke, and just two days before she gave birth. </p><p>In another, a manager’s own performance review noted that his demotion followed his return from medical leave; he was put on the list sixteen days into a second medical leave.</p><p>The 26 plaintiffs claim that Meta has violated state-protected leave laws, the Family and Medical Leave Act, the Pregnancy Discrimination Act, and the Americans with Disabilities Act.</p><p>They're calling for a preliminary court ruling that would stop Meta from finalizing the layoffs while they pursue their ​claims.</p><p>“Once these separations are final, the harms are irreversible: employer-subsidized health coverage lost during pregnancy, postpartum recovery, and active medical treatment; time-bound leave rights extinguished; unvested equity forfeited; and immigration consequences triggered,” they said.</p><p>Plaintiffs are also seeking financial compensation that could include reinstatement, back pay, lost equity, benefits, and other damages.</p><h2 id="ai-in-hr-is-a-delicate-balancing-act">AI in HR is a delicate balancing act</h2><p>The use of AI systems in HR and recruitment has been a long-running point of contention, particularly with regard to the <a href="https://www.itpro.com/technology/artificial-intelligence/will-ai-hiring-entrench-gender-bias">potential for bias or discrimination</a>. </p><p>Workday, for example, is facing a class-action lawsuit amidst claims its AI software discriminated against job applicants. As <a href="https://www.itpro.com/business/policy-legislation/370133/workday-hit-with-claims-its-ai-hiring-systems-are-discriminatory"><em>ITPro </em>reported in June</a>, the company hit back at the claims, with a spokesperson insisting its AI solutions “don’t make hiring decisions”. </p><p>Ilia Kolochenko, founder of cybersecurity company ImmuniWeb and a lawyer practising in <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>and data protection, noted that over-regulation of the use of AI in HR could cause long-term harm. </p><p>"With the current trend to restrict or even entirely ban AI in HR decision-making processes, most organizations will either conceal the use of AI or shift back to non-AI systems especially in those jurisdictions that have no GDPR-like protection against automated decision-making on human subjects," he said.</p><p>"Compared to AI-powered HR systems, their non-AI homologues are quite primitive, fail to consider the relevant context and often provide incorrect metrics or data,” Kolochenko added. </p><p>“For instance, a poor performance of a delivery truck driver will not be correlated with extreme summer heat or winter snowfalls, sudden family loss or simply a technical issue with the vehicle. Eventually, innocent persons are wrongly punished.”</p><p><em>ITPro </em>approached Meta for comment, but did not receive a response by time of publication.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/business-strategy/meta-used-a-constellation-of-internal-artificial-intelligence-systems-to-target-workers-in-recent-layoffs-lawsuit-claims-keystroke-monitoring-data-ai-token-usage-and-performance-ratings-allegedly-decided-employee-fates</link>
                                                                            <description>
                            <![CDATA[ Former Meta employees allege that an AI system used to select people for layoffs was biased ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">9smGFa5EcmxejWYs9ovEP7</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/HxRbEb62C8beryHZJQcsVQ-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 15 Jul 2026 10:54:33 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:50 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Business Strategy]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/HxRbEb62C8beryHZJQcsVQ-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Meta logo and branding pictured on a sign at the entrance to the company&#039;s global headquarters at One Hacker Way in Menlo Park, California.]]></media:description>                                                            <media:text><![CDATA[Meta logo and branding pictured on a sign at the entrance to the company&#039;s global headquarters at One Hacker Way in Menlo Park, California.]]></media:text>
                                <media:title type="plain"><![CDATA[Meta logo and branding pictured on a sign at the entrance to the company&#039;s global headquarters at One Hacker Way in Menlo Park, California.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/HxRbEb62C8beryHZJQcsVQ-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>A group of current and former Meta employees is suing the company, claiming that it used AI to target those on medical or family leave for layoffs.</p><p>In May, Meta announced it was <a href="https://www.itpro.com/business/leadership/internal-memo-suggests-meta-will-lay-off-10-percent-of-its-employees-with-a-further-10-percent-set-to-be-transferred-to-better-focus-on-ai"><u>cutting around 8,000 staff</u></a>, representing roughly 10% of its workforce. According to the lawsuit, the tech giant used AI systems to identify those set for redundancy. </p><p>These systems allegedly took into account keystroke monitoring data, AI token usage </p><p>“Meta did not assemble the termination list through the considered judgment of managers who knew the work,” <a href="https://storage.courtlistener.com/recap/gov.uscourts.cand.474171/gov.uscourts.cand.474171.1.0.pdf" target="_blank">court filings</a> state. </p><p>“Instead, Meta used a constellation of internal artificial intelligence systems — including a system referred to internally as ‘Metamate,’ employee-trained ‘second-brain’ agents, keystroke- and activity-monitoring data, AI-token-usage dashboards, and algorithmically assisted performance ranking and calibration — to score, rank and select employees for inclusion on the list.”</p><p>Complainants allege this put employees on protected medical or family leave, or those whose output was reduced by a disability, at a disadvantage by penalizing them with reduced performance scores. </p><p>"The result was that employees who took protected leaves were disproportionately selected for layoff, based on scoring that not only failed to account for their protected leaves, but in effect penalized the employees for exercising their legal rights to these leaves."</p><p>In one example, an employee was laid off while on approved pre-birth pregnancy leave – the day before her waters broke, and just two days before she gave birth. </p><p>In another, a manager’s own performance review noted that his demotion followed his return from medical leave; he was put on the list sixteen days into a second medical leave.</p><p>The 26 plaintiffs claim that Meta has violated state-protected leave laws, the Family and Medical Leave Act, the Pregnancy Discrimination Act, and the Americans with Disabilities Act.</p><p>They're calling for a preliminary court ruling that would stop Meta from finalizing the layoffs while they pursue their ​claims.</p><p>“Once these separations are final, the harms are irreversible: employer-subsidized health coverage lost during pregnancy, postpartum recovery, and active medical treatment; time-bound leave rights extinguished; unvested equity forfeited; and immigration consequences triggered,” they said.</p><p>Plaintiffs are also seeking financial compensation that could include reinstatement, back pay, lost equity, benefits, and other damages.</p><h2 id="ai-in-hr-is-a-delicate-balancing-act">AI in HR is a delicate balancing act</h2><p>The use of AI systems in HR and recruitment has been a long-running point of contention, particularly with regard to the <a href="https://www.itpro.com/technology/artificial-intelligence/will-ai-hiring-entrench-gender-bias">potential for bias or discrimination</a>. </p><p>Workday, for example, is facing a class-action lawsuit amidst claims its AI software discriminated against job applicants. As <a href="https://www.itpro.com/business/policy-legislation/370133/workday-hit-with-claims-its-ai-hiring-systems-are-discriminatory"><em>ITPro </em>reported in June</a>, the company hit back at the claims, with a spokesperson insisting its AI solutions “don’t make hiring decisions”. </p><p>Ilia Kolochenko, founder of cybersecurity company ImmuniWeb and a lawyer practising in <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>and data protection, noted that over-regulation of the use of AI in HR could cause long-term harm. </p><p>"With the current trend to restrict or even entirely ban AI in HR decision-making processes, most organizations will either conceal the use of AI or shift back to non-AI systems especially in those jurisdictions that have no GDPR-like protection against automated decision-making on human subjects," he said.</p><p>"Compared to AI-powered HR systems, their non-AI homologues are quite primitive, fail to consider the relevant context and often provide incorrect metrics or data,” Kolochenko added. </p><p>“For instance, a poor performance of a delivery truck driver will not be correlated with extreme summer heat or winter snowfalls, sudden family loss or simply a technical issue with the vehicle. Eventually, innocent persons are wrongly punished.”</p><p><em>ITPro </em>approached Meta for comment, but did not receive a response by time of publication.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Understanding the hidden costs of sovereign AI ]]></title>
                                                                                                <dc:content><![CDATA[ <p><a href="https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech">Sovereign AI</a> has quickly moved from policy language into boardroom planning. For CIOs, the appeal is easy to understand: more control over where data goes, which systems can access it, and how far an organization depends on overseas providers for increasingly critical AI workloads.</p><p>In 2026, the pressure is only growing. Governments want national AI capacity, regulators want clearer accountability, and enterprises want to know whether their data, models, and infrastructure are exposed to legal or operational risks that they cannot easily see.</p><p>In that context, regional hosting, <a href="https://www.itpro.com/cloud/cloud-computing/what-is-a-sovereign-cloud">sovereign cloud services</a>, and local <a href="https://www.itpro.com/infrastructure/ai-infrastructure-global-divide">AI infrastructure</a> can all look like a sensible response.</p><p>Gartner has <a href="https://www.gartner.com/en/newsroom/press-releases/2026-01-29-gartner-predicts-35-percent-of-countries-will-be-locked-into-region-specific-ai-platforms-by-2027" target="_blank">predicted </a>that 35% of countries will be locked into <a href="https://www.itpro.com/technology/artificial-intelligence/eu-businesses-will-flock-to-region-specific-ai-platforms-by-2027-but-cost-could-be-a-major-hurdle">region-specific AI platforms by 2027</a>, up from 5% in 2026, underlining how quickly sovereignty is becoming a question of procurement, platform choice, and long-term flexibility.</p><p>But sovereignty is not a single switch that makes an AI deployment safer, simpler, or more compliant, and can actually narrow model choice, add procurement friction, raise costs, complicate integration, and create fresh dependencies. </p><p>The real question for IT leaders is what level of control each workload actually needs, what the organization is prepared to give up in return, and whether a sovereign approach reduces risk or simply moves it somewhere less obvious, not whether sovereign AI is “good or bad”.</p><p>To help understand these trade-offs, <em>ITPro </em>spoke to Forrester senior analyst <a href="https://www.forrester.com/analyst-bio/dario-maisto/BIO20009" target="_blank"><u>Dario Maisto</u></a> about what enterprises really mean when they talk about “sovereign AI”, where local control can reduce risk, and where it can create fresh complexity.</p><h2 id="what-enterprises-really-mean-by-sovereign-ai">What enterprises really mean by “sovereign AI”</h2><p>For many businesses, sovereign AI still starts with a fairly narrow concern: keeping data in the right place, which might mean running inference in a specific region, storing prompts and outputs locally, or making sure sensitive data is processed under a particular legal regime.</p><p>While these are important controls, especially for regulated sectors, they are only one part of the sovereignty picture. </p><p>Maisto says most enterprise projects have not yet reached the point where organizations are wrestling with sovereign AI at full production scale. “Clients' journeys are not that advanced in AI deployments that they can worry about the depth of sovereign AI at scale for production environments,” he says. “They have mostly data residency concerns.”</p><p>Maisto’s insight explains why the term can become slippery. A vendor might describe an AI service as “sovereign” because the data is hosted in-country; a public sector buyer might use the same term to mean local legal accountability or domestic infrastructure; and so on. </p><p>Distinctions like this are important because AI does not sit neatly in one place. A single deployment can involve cloud infrastructure, foundation models, APIs, orchestration tools, identity systems, logging, monitoring, and a lot more besides. </p><p>Once AI agents – perhaps 2026’s buzziest trend – enter the picture, the stack becomes even harder to contain, as systems start pulling from enterprise data sources and triggering actions across multiple applications.</p><h2 id="why-local-hosting-is-not-the-same-as-control">Why local hosting is not the same as control</h2><p>The first trap is assuming that sovereignty begins and ends with location. Keeping data in a specific country or region can be important, but local hosting does not automatically decide who controls the service, which laws apply, or who can access the underlying systems.</p><p>Maisto is blunt on this point: “Local hosting does not protect workloads and data with regard to sovereignty concerns. This is where organizations overestimate the potential of local hosting and in-region infrastructure to improve their digital sovereignty posture.”</p><p>“If the infrastructure and the tools are owned by a third party operating under a different jurisdiction, the sovereignty risk is not mitigated,” he says. </p><p>None of these makes regional hosting pointless. It can still help with latency, regulatory alignment, audit requirements, and internal assurance, and can also reduce the number of cross-border transfers and make it easier to show that sensitive workloads are being handled within a defined legal environment.</p><p>The problem comes when residency is treated as a substitute for sovereignty. For CIOs, the due diligence needs to go beyond the region on the invoice and into the architecture behind the service.</p><h2 id="sovereignty-can-limit-choice">Sovereignty can limit choice</h2><p>The next cost to businesses is flexibility. AI buyers are used to a market that moves quickly (perhaps too quickly), with new models, tools, and managed services appearing every few months or even weeks.</p><p>A sovereign AI strategy can slow that down, especially if an organization has to use approved regions, certified providers, local infrastructure, or a smaller set of compliant services, alongside additional procurement, legal, or regulatory scrutiny.</p><p>While that trade-off may be perfectly reasonable for the right workload, it still needs to be visible. A government department handling citizen data should not treat model access like a marketing team testing copy variations, and a bank, healthcare provider, or critical infrastructure operator may decide that tighter controls matter more than immediate access to the newest frontier model.</p><p>But a sovereign deployment may mean fewer models, less access to specialist AI services, slower feature rollouts, or more integration work for tools that would otherwise be available through a major cloud platform.</p><p>That risk is already visible in agentic AI, where <em>ITPro </em>has <a href="https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech"><u>reported</u></a> that many UK companies are moving ahead with deployments despite gaps in governance and visibility over where data is stored, processed, and accessed.</p><p>Maisto argues this is where buyers can underestimate the knock-on effects. “Organizations mostly tackle this theme from a data residency and inference location perspective, although problems like agents' sovereignty could be more important, and the solutions look very immature as of now in the market,” he says.   </p><p>“Integration is another underestimated aspect of agentic AI as agents are pervasive and have the potential to increase vendor lock-in with non-sovereign vendors.”</p><p>As AI becomes more embedded in enterprise systems, sovereignty becomes harder to separate from day-to-day architecture. The key question is how deeply it connects to the business, and whether those connections can still be governed, audited, or replaced.</p><h2 id="skills-and-procurement-can-slow-everything-down">Skills and procurement can slow everything down</h2><p>Even when the case for sovereign AI is clear, delivery can be harder than the strategy suggests. An organization still needs people who can assess suppliers, design the architecture, check operational controls, and so on, and these skills are already scarce in mainstream AI projects, before sovereignty requirements narrow the pool further.</p><p>Talent may become one of the biggest constraints, according to Maisto. “According to our forecast, [the] tech workforce is the only sovereignty index that is going to decrease in the next five years,” he adds. </p><p>“The more specialized the skills, the less the chance to have the right talent with the needed sovereignty requirements at scale. With an outlook to the future, skills from a sovereign pool of talent would be the most relevant component.”</p><p>Procurement adds another layer of friction. </p><p>Sovereign AI buyers may need to test where data is stored, how support access works, and many other features of the wider operational chain. Most of the time, the process is likely to be slower than buying a standard AI service from an existing cloud marketplace.</p><p>For CIOs, this creates a now-familiar tension: The business wants access to AI quickly, while security, legal, and compliance teams need confidence that the deployment will not create exposure later. </p><p>Sovereign AI can reduce some risks, but only if the organization has the skills and procurement discipline to define what “sovereign” means before it buys.</p><h2 id="should-companies-prioritize-sovereign-ai">Should companies prioritize sovereign AI? </h2><p>The safest approach is to treat sovereign AI as a design choice with a defined purpose.</p><p>Maisto says Forrester uses a “<a href="https://www.forrester.com/blogs/minimum-viable-sovereignty-a-smarter-path-for-tech-leaders/"><u>Minimum Viable Sovereignty</u></a> Model Decision Framework” to help technology leaders weigh “compliance requirements, risk appetite, budget, tech and business requirements, and available sovereign products and services.”</p><p>He adds that they should assess “the tradeoff between sovereignty and functionality, looking at the viability, feasibility and desirability of a sovereign solution.”</p><p>Any assessment should include the practical value of sovereignty, the compromises it introduces, and the maturity of the products available. “Not all vendors are scoring equally on AI capabilities and even less so when it comes to sovereign AI capabilities,” says Maisto. </p><p>For CIOs, the aim should be proportional control. Some workloads will justify strict sovereignty requirements because the data, operational risk, or public accountability demands it; others may be better served by strong governance, clear deployment controls, and encryption. </p><p>The danger comes when enterprises buy sovereign AI too vaguely, without defining which risks they are trying to reduce, which freedoms they are willing to give up, and what they will need back if the strategy changes.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/the-hidden-cost-of-sovereign-ai-what-control-really-buys-you-and-what-it-breaks</link>
                                                                            <description>
                            <![CDATA[ While sovereign AI could help bolster data protection and privacy, it's could create fresh dependencies, increase costs, and narrow model choice ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">HoPCGTu6iTR7EgsEYzSgMY</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/3MYdgoHSru8pFz22e58jtQ-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 15 Jul 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:35 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Max Slater-Robins ]]></dc:creator>                                                                                                        <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/3MYdgoHSru8pFz22e58jtQ-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Futuristic design of artificial Intelligence brain with circuit board.]]></media:description>                                                            <media:text><![CDATA[Futuristic design of artificial Intelligence brain with circuit board.]]></media:text>
                                <media:title type="plain"><![CDATA[Futuristic design of artificial Intelligence brain with circuit board.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/3MYdgoHSru8pFz22e58jtQ-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p><a href="https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech">Sovereign AI</a> has quickly moved from policy language into boardroom planning. For CIOs, the appeal is easy to understand: more control over where data goes, which systems can access it, and how far an organization depends on overseas providers for increasingly critical AI workloads.</p><p>In 2026, the pressure is only growing. Governments want national AI capacity, regulators want clearer accountability, and enterprises want to know whether their data, models, and infrastructure are exposed to legal or operational risks that they cannot easily see.</p><p>In that context, regional hosting, <a href="https://www.itpro.com/cloud/cloud-computing/what-is-a-sovereign-cloud">sovereign cloud services</a>, and local <a href="https://www.itpro.com/infrastructure/ai-infrastructure-global-divide">AI infrastructure</a> can all look like a sensible response.</p><p>Gartner has <a href="https://www.gartner.com/en/newsroom/press-releases/2026-01-29-gartner-predicts-35-percent-of-countries-will-be-locked-into-region-specific-ai-platforms-by-2027" target="_blank">predicted </a>that 35% of countries will be locked into <a href="https://www.itpro.com/technology/artificial-intelligence/eu-businesses-will-flock-to-region-specific-ai-platforms-by-2027-but-cost-could-be-a-major-hurdle">region-specific AI platforms by 2027</a>, up from 5% in 2026, underlining how quickly sovereignty is becoming a question of procurement, platform choice, and long-term flexibility.</p><p>But sovereignty is not a single switch that makes an AI deployment safer, simpler, or more compliant, and can actually narrow model choice, add procurement friction, raise costs, complicate integration, and create fresh dependencies. </p><p>The real question for IT leaders is what level of control each workload actually needs, what the organization is prepared to give up in return, and whether a sovereign approach reduces risk or simply moves it somewhere less obvious, not whether sovereign AI is “good or bad”.</p><p>To help understand these trade-offs, <em>ITPro </em>spoke to Forrester senior analyst <a href="https://www.forrester.com/analyst-bio/dario-maisto/BIO20009" target="_blank"><u>Dario Maisto</u></a> about what enterprises really mean when they talk about “sovereign AI”, where local control can reduce risk, and where it can create fresh complexity.</p><h2 id="what-enterprises-really-mean-by-sovereign-ai">What enterprises really mean by “sovereign AI”</h2><p>For many businesses, sovereign AI still starts with a fairly narrow concern: keeping data in the right place, which might mean running inference in a specific region, storing prompts and outputs locally, or making sure sensitive data is processed under a particular legal regime.</p><p>While these are important controls, especially for regulated sectors, they are only one part of the sovereignty picture. </p><p>Maisto says most enterprise projects have not yet reached the point where organizations are wrestling with sovereign AI at full production scale. “Clients' journeys are not that advanced in AI deployments that they can worry about the depth of sovereign AI at scale for production environments,” he says. “They have mostly data residency concerns.”</p><p>Maisto’s insight explains why the term can become slippery. A vendor might describe an AI service as “sovereign” because the data is hosted in-country; a public sector buyer might use the same term to mean local legal accountability or domestic infrastructure; and so on. </p><p>Distinctions like this are important because AI does not sit neatly in one place. A single deployment can involve cloud infrastructure, foundation models, APIs, orchestration tools, identity systems, logging, monitoring, and a lot more besides. </p><p>Once AI agents – perhaps 2026’s buzziest trend – enter the picture, the stack becomes even harder to contain, as systems start pulling from enterprise data sources and triggering actions across multiple applications.</p><h2 id="why-local-hosting-is-not-the-same-as-control">Why local hosting is not the same as control</h2><p>The first trap is assuming that sovereignty begins and ends with location. Keeping data in a specific country or region can be important, but local hosting does not automatically decide who controls the service, which laws apply, or who can access the underlying systems.</p><p>Maisto is blunt on this point: “Local hosting does not protect workloads and data with regard to sovereignty concerns. This is where organizations overestimate the potential of local hosting and in-region infrastructure to improve their digital sovereignty posture.”</p><p>“If the infrastructure and the tools are owned by a third party operating under a different jurisdiction, the sovereignty risk is not mitigated,” he says. </p><p>None of these makes regional hosting pointless. It can still help with latency, regulatory alignment, audit requirements, and internal assurance, and can also reduce the number of cross-border transfers and make it easier to show that sensitive workloads are being handled within a defined legal environment.</p><p>The problem comes when residency is treated as a substitute for sovereignty. For CIOs, the due diligence needs to go beyond the region on the invoice and into the architecture behind the service.</p><h2 id="sovereignty-can-limit-choice">Sovereignty can limit choice</h2><p>The next cost to businesses is flexibility. AI buyers are used to a market that moves quickly (perhaps too quickly), with new models, tools, and managed services appearing every few months or even weeks.</p><p>A sovereign AI strategy can slow that down, especially if an organization has to use approved regions, certified providers, local infrastructure, or a smaller set of compliant services, alongside additional procurement, legal, or regulatory scrutiny.</p><p>While that trade-off may be perfectly reasonable for the right workload, it still needs to be visible. A government department handling citizen data should not treat model access like a marketing team testing copy variations, and a bank, healthcare provider, or critical infrastructure operator may decide that tighter controls matter more than immediate access to the newest frontier model.</p><p>But a sovereign deployment may mean fewer models, less access to specialist AI services, slower feature rollouts, or more integration work for tools that would otherwise be available through a major cloud platform.</p><p>That risk is already visible in agentic AI, where <em>ITPro </em>has <a href="https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech"><u>reported</u></a> that many UK companies are moving ahead with deployments despite gaps in governance and visibility over where data is stored, processed, and accessed.</p><p>Maisto argues this is where buyers can underestimate the knock-on effects. “Organizations mostly tackle this theme from a data residency and inference location perspective, although problems like agents' sovereignty could be more important, and the solutions look very immature as of now in the market,” he says.   </p><p>“Integration is another underestimated aspect of agentic AI as agents are pervasive and have the potential to increase vendor lock-in with non-sovereign vendors.”</p><p>As AI becomes more embedded in enterprise systems, sovereignty becomes harder to separate from day-to-day architecture. The key question is how deeply it connects to the business, and whether those connections can still be governed, audited, or replaced.</p><h2 id="skills-and-procurement-can-slow-everything-down">Skills and procurement can slow everything down</h2><p>Even when the case for sovereign AI is clear, delivery can be harder than the strategy suggests. An organization still needs people who can assess suppliers, design the architecture, check operational controls, and so on, and these skills are already scarce in mainstream AI projects, before sovereignty requirements narrow the pool further.</p><p>Talent may become one of the biggest constraints, according to Maisto. “According to our forecast, [the] tech workforce is the only sovereignty index that is going to decrease in the next five years,” he adds. </p><p>“The more specialized the skills, the less the chance to have the right talent with the needed sovereignty requirements at scale. With an outlook to the future, skills from a sovereign pool of talent would be the most relevant component.”</p><p>Procurement adds another layer of friction. </p><p>Sovereign AI buyers may need to test where data is stored, how support access works, and many other features of the wider operational chain. Most of the time, the process is likely to be slower than buying a standard AI service from an existing cloud marketplace.</p><p>For CIOs, this creates a now-familiar tension: The business wants access to AI quickly, while security, legal, and compliance teams need confidence that the deployment will not create exposure later. </p><p>Sovereign AI can reduce some risks, but only if the organization has the skills and procurement discipline to define what “sovereign” means before it buys.</p><h2 id="should-companies-prioritize-sovereign-ai">Should companies prioritize sovereign AI? </h2><p>The safest approach is to treat sovereign AI as a design choice with a defined purpose.</p><p>Maisto says Forrester uses a “<a href="https://www.forrester.com/blogs/minimum-viable-sovereignty-a-smarter-path-for-tech-leaders/"><u>Minimum Viable Sovereignty</u></a> Model Decision Framework” to help technology leaders weigh “compliance requirements, risk appetite, budget, tech and business requirements, and available sovereign products and services.”</p><p>He adds that they should assess “the tradeoff between sovereignty and functionality, looking at the viability, feasibility and desirability of a sovereign solution.”</p><p>Any assessment should include the practical value of sovereignty, the compromises it introduces, and the maturity of the products available. “Not all vendors are scoring equally on AI capabilities and even less so when it comes to sovereign AI capabilities,” says Maisto. </p><p>For CIOs, the aim should be proportional control. Some workloads will justify strict sovereignty requirements because the data, operational risk, or public accountability demands it; others may be better served by strong governance, clear deployment controls, and encryption. </p><p>The danger comes when enterprises buy sovereign AI too vaguely, without defining which risks they are trying to reduce, which freedoms they are willing to give up, and what they will need back if the strategy changes.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Microsoft CEO Satya Nadella says enterprises shouldn’t be sharing so much data with AI providers ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Companies using AI should be wary about losing control of their data – and the risk of AI developers making use of that knowledge. </p><p>That’s according to <a href="https://www.itpro.com/technology/artificial-intelligence/satya-nadella-microsoft-ai-slop-2026">Microsoft CEO Satya Nadella</a>, who in a recent <a href="https://snscratchpad.com/posts/reverse-information-paradox/" target="_blank"><u>blog post</u></a> warned many AI users are "paying twice" thanks to what he calls the "reverse information paradox”.</p><p>This is a process through which AI providers, particularly frontier labs, get access to proprietary data about their own customers. Nadella said the trend represents a huge risk for enterprises, and called for protections akin to patents. </p><p>The critique is intriguing as Microsoft itself is an AI company, offering plenty of AI-driven products to customers and cramming AI features into its software. Beyond that, Microsoft was an early funder of OpenAI, though <a href="https://www.itpro.com/technology/artificial-intelligence/the-honeymoon-period-is-officially-over-for-microsoft-and-openai">that relationship has weakened of late</a>. </p><p>Nadella pointed to economist Kenneth Arrow's information paradox, in which the seller of data or knowledge risks giving it away in order to prove it's worth selling. He noted that AI has created the reverse problem, in which the buyer gives away knowledge in order to use the product purchased. </p><p>"You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful," he noted. "The better you want the model to perform, the more of that knowledge you have to feed it!"</p><p>"Over time, the information asymmetry becomes increasingly skewed. The seller learns more and more about you as you use what you purchased, while you learn very little about what the seller is learning in return."</p><h2 id="the-ai-exhaust">The ‘AI exhaust’</h2><p>This concern isn't just about uploading corporate data to systems, or accidentally including proprietary data in a prompt, though those are both concerns. Instead, Nadella said every interaction with AI inherently includes information that could be useful. </p><p>"Models learn from 'exhaust,' the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong," he wrote. </p><p>"Every correction is distilled into institutional know-how. It’s the kind of knowledge a competitor could never buy, and the kind that leaks almost imperceptibly: trace by trace, correction by correction, eval by eval."</p><p>Ironically, Nadella said it was necessary for the "great innovation" of allowing model developers to use public data – some of them face lawsuits for hoovering up copyrighted data to train their systems. </p><p>When it comes to business data, providers should share anything learned with customers. Very few actually do this, however. </p><p>"If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself," he said. "Therefore, it’s imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop."</p><h2 id="nadella-s-tips-on-how-to-protect-your-company-from-ai">Nadella's tips on how to protect your company from AI</h2><p>Nadella said the existing setup requires a protection system on par with patents, but also advised companies to set a "trust boundary" by considering a few key aspects – some of which may be easier by simply switching to <a href="https://www.itpro.com/software/open-source/the-pros-and-cons-of-open-source-ai-for-business"><u>open source AI</u></a>. </p><p>He suggested companies should control their own "evals", or evaluations of the AI system, because that reveals what looks "good" to the organization, and keep ownership of decisions, feedbacks,  and other outputs for their own use. </p><p>Beyond that, companies should build capability with AI via their own learning environments where models can learn using real workflows without exposing corporate data. </p><p>Improving choice can be achieved by decoupling the orchestration layer from any single model, enabling the ability to switch to another model if one is withdrawn. This can also help organizations cut costs by choosing the right model based on their individual needs. </p><p>Lastly, he said to "compound" those four ideas to "create your own continuous learning loop". </p><p>"In other words, a company should be able to use a model without giving up the knowledge that makes it unique," he added. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/a-company-should-be-able-to-use-a-model-without-giving-up-the-knowledge-that-makes-it-unique-microsoft-ceo-satya-nadella-says-enterprises-shouldnt-be-sharing-so-much-data-with-ai-providers</link>
                                                                            <description>
                            <![CDATA[ The Microsoft chief warned that corporate data could be at risk thanks to AI models ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">AJcEqjeYAFnVqEW95ovKbc</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/3impDCwWStvn4ovtBBf9o7-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 14 Jul 2026 10:32:47 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:24 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/3impDCwWStvn4ovtBBf9o7-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Microsoft CEO Satya Nadella pictured on stage at the Microsoft 50th Anniversary event in Redmond, Washington, with the company logo on a screen to his right.]]></media:description>                                                            <media:text><![CDATA[Microsoft CEO Satya Nadella pictured on stage at the Microsoft 50th Anniversary event in Redmond, Washington, with the company logo on a screen to his right.]]></media:text>
                                <media:title type="plain"><![CDATA[Microsoft CEO Satya Nadella pictured on stage at the Microsoft 50th Anniversary event in Redmond, Washington, with the company logo on a screen to his right.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/3impDCwWStvn4ovtBBf9o7-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Companies using AI should be wary about losing control of their data – and the risk of AI developers making use of that knowledge. </p><p>That’s according to <a href="https://www.itpro.com/technology/artificial-intelligence/satya-nadella-microsoft-ai-slop-2026">Microsoft CEO Satya Nadella</a>, who in a recent <a href="https://snscratchpad.com/posts/reverse-information-paradox/" target="_blank"><u>blog post</u></a> warned many AI users are "paying twice" thanks to what he calls the "reverse information paradox”.</p><p>This is a process through which AI providers, particularly frontier labs, get access to proprietary data about their own customers. Nadella said the trend represents a huge risk for enterprises, and called for protections akin to patents. </p><p>The critique is intriguing as Microsoft itself is an AI company, offering plenty of AI-driven products to customers and cramming AI features into its software. Beyond that, Microsoft was an early funder of OpenAI, though <a href="https://www.itpro.com/technology/artificial-intelligence/the-honeymoon-period-is-officially-over-for-microsoft-and-openai">that relationship has weakened of late</a>. </p><p>Nadella pointed to economist Kenneth Arrow's information paradox, in which the seller of data or knowledge risks giving it away in order to prove it's worth selling. He noted that AI has created the reverse problem, in which the buyer gives away knowledge in order to use the product purchased. </p><p>"You essentially pay for intelligence twice, once with money, and again with something even more valuable: the proprietary knowledge you must reveal to make that intelligence useful," he noted. "The better you want the model to perform, the more of that knowledge you have to feed it!"</p><p>"Over time, the information asymmetry becomes increasingly skewed. The seller learns more and more about you as you use what you purchased, while you learn very little about what the seller is learning in return."</p><h2 id="the-ai-exhaust">The ‘AI exhaust’</h2><p>This concern isn't just about uploading corporate data to systems, or accidentally including proprietary data in a prompt, though those are both concerns. Instead, Nadella said every interaction with AI inherently includes information that could be useful. </p><p>"Models learn from 'exhaust,' the prompts people write, the tools agents use, and especially the corrections people make when the model is wrong," he wrote. </p><p>"Every correction is distilled into institutional know-how. It’s the kind of knowledge a competitor could never buy, and the kind that leaks almost imperceptibly: trace by trace, correction by correction, eval by eval."</p><p>Ironically, Nadella said it was necessary for the "great innovation" of allowing model developers to use public data – some of them face lawsuits for hoovering up copyrighted data to train their systems. </p><p>When it comes to business data, providers should share anything learned with customers. Very few actually do this, however. </p><p>"If learning flows in only one direction, economic value converges toward the owners of the learning infrastructure rather than the creators of the knowledge itself," he said. "Therefore, it’s imperative that we distribute the learning infrastructure to every firm so that they can control their own learning loop."</p><h2 id="nadella-s-tips-on-how-to-protect-your-company-from-ai">Nadella's tips on how to protect your company from AI</h2><p>Nadella said the existing setup requires a protection system on par with patents, but also advised companies to set a "trust boundary" by considering a few key aspects – some of which may be easier by simply switching to <a href="https://www.itpro.com/software/open-source/the-pros-and-cons-of-open-source-ai-for-business"><u>open source AI</u></a>. </p><p>He suggested companies should control their own "evals", or evaluations of the AI system, because that reveals what looks "good" to the organization, and keep ownership of decisions, feedbacks,  and other outputs for their own use. </p><p>Beyond that, companies should build capability with AI via their own learning environments where models can learn using real workflows without exposing corporate data. </p><p>Improving choice can be achieved by decoupling the orchestration layer from any single model, enabling the ability to switch to another model if one is withdrawn. This can also help organizations cut costs by choosing the right model based on their individual needs. </p><p>Lastly, he said to "compound" those four ideas to "create your own continuous learning loop". </p><p>"In other words, a company should be able to use a model without giving up the knowledge that makes it unique," he added. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ FCA recommends expanded powers to boost financial services AI regulation ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Stronger powers could be needed to regulate the use of AI in financial services as consumers flock to the technology for advice, according to the Financial Conduct Authority (FCA).</p><p>Following an investigation into the technology and its impact on consumer finance activities, the regulator has recommended a formal review into how to regulate the use of general-purpose LLMs across the industry. </p><p>The FCA report – known as the Mills Review after its author – noted that companies and consumers are increasingly delegating decisions about finances to AI and AI-powered systems. </p><p>A survey by the regulator found that one-in-five adults were already open to the idea of general-purpose <a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> – such as Claude, <a href="https://www.itpro.com/technology/artificial-intelligence/openai-chatgpt-superapp-overhaul-public-listing">ChatGPT</a>, or Gemini – making financial decisions on their behalf, in particular around debt advice, pensions, and investments. </p><p>While that could help reduce existing challenges such as "advice gaps" and friction around switching services, the FCA warned about risks including fraud and scams, as well as disrupting market structure and competition. </p><p>"AI offers a once-in-a-generation chance to close the information asymmetries and frictions that have long left people making poor financial decisions," said FCA executive director and report author Sheldon Mills in the forward to the review. </p><p>"The right spur, in retail financial services, is ensuring consumers can make healthy ones; regulation, whether supportive or restrictive, should serve that outcome."</p><p>The FCA noted that many consumers may not be aware that they have no formal recourse if something goes wrong, and said some assume they receive equal protection akin to traditional sources of advice. </p><p>Emma Banymandhub, CEO of The Payments Association, agreed that the gap in understanding could cause issues and urged caution on the part of consumers. </p><p>"Consumers may be increasingly comfortable using AI agents for routine tasks such as weekly shopping, but AI-driven savings and investment decisions present a very different set of challenges," Banymandhub commented. </p><p>"While AI can increasingly explain investment concepts and analyse financial information, personalised investment recommendations remain subject to important regulatory constraints."</p><h2 id="ai-gains-traction-in-financial-services">AI gains traction in financial services</h2><p>The financial services sector has rapidly emerged as one of the key growth spaces in terms of the use of AI, with a host of major providers across the UK ramping up adoption of the technology. </p><p>As <em>ITPro </em>previously reported, high street lenders such as <a href="https://www.itpro.com/technology/artificial-intelligence/using-generative-ai-as-a-copilot-is-the-sweet-spot-a-look-at-nationwides-ai-approach">Nationwide</a>, Lloyds Banking Group, and <a href="https://www.itpro.com/business/business-strategy/yorkshire-building-society-touts-customer-service-gains-with-ai-agents">Yorkshire Building Society</a> have all made significant strides on this front over the last six months. </p><p>A <a href="https://publications.parliament.uk/pa/cm5901/cmselect/cmtreasy/684/report.html" target="_blank"><u>parliamentary inquiry</u></a> into the use of AI in financial services, published in January 2026, found that the sector “substantially outpaces” others with regard to AI adoption. </p><p>Indeed, around 75% of UK-based financial services firms are now using the technology, with insurers and large banks among the most aggressive in their pursuit of AI adoption. </p><p>Running parallel to this, the use of AI by consumers is also surging, according to research from Lloyds Banking Group. The firm’s 2025 <a href="https://www.lloydsbankinggroup.com/media/press-releases/2025/lloyds-banking-group-2025/28m-adults-using-ai-to-manage-money.html" target="_blank"><u><em>Consumer Digital Index</em></u></a> found that AI has “rapidly become a financial tool for millions across the UK". </p><p>56% of adults – equivalent to around 28 million people – revealed they’d used AI over the preceding 12 months for financial advice. ChatGPT, for example, was referenced as the most popular platform in this regard, used by six-in-10 consumers. </p><h2 id="what-the-mills-review-recommends">What the Mills Review recommends</h2><p>With this in mind, the Mills Review recommended that the existing "regulatory perimeter" needs to be secured and adapted to take in AI's impact on retail financial services – and that should come via a review within the next six months into how AI is affecting the market. </p><p>"The review should examine how consumers use AI including general purpose LLM tools for personal financial management across savings, investments, pensions, mortgages and debt management, and the implications for competition, innovation and growth," the Mills Review said. </p><p>"It should examine the risks of consumer harm, including how far its usage has or will move along the autonomy spectrum, and any impacts on market integrity and the potential for regulatory arbitrage."</p><p>Based on that review's guidance, the FCA could tweak regulation as necessary, the report said. Beyond that, the review called for the FCA to monitor for evidence of harm and new consumer models, engage with providers for better insight into changes, and be ready with intervention tools and mechanisms as necessary. </p><h2 id="frontier-model-monitoring-in-finance">Frontier model monitoring in finance</h2><p>In the longer term, the Mills Review called for the FCA to keep an eye on frontier model capabilities and AI adoption, considering its impact on its regulatory work, in particular where use of AI falls outside of existing protections and once AI agents start taking more action on behalf of consumers. </p><p>The review also called for stronger powers for the FCA so it can look at these issues more widely. </p><p>"As is clear in the report, we need to keep pace with a rapidly changing environment and the principles-based, outcomes focussed approach we’ve taken on AI – relying on the Consumer Duty and Senior Managers Regime – has been critical to us doing so," said Ashley Alder, chair of the FCA. </p><p>"The recommendations build on work the FCA has been doing – not least allowing firms to test their use of AI with us – and our own use of AI to be a smarter regulator, more efficient and effective."</p><p>Banymandhub<strong> </strong>added that walking that balance between enabling AI and protecting consumers and companies was key.  </p><p>"The FCA’s Mills Review reinforces that firms should treat agentic AI as an accountability and governance issue now, while providing greater confidence to innovate responsibly as AI adoption accelerates," she said. </p><p>"AI has enormous potential for financial services, but realising that potential will depend on strong governance, clear accountability and maintaining consumer trust."</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/policy-and-legislation/fca-recommends-expanded-powers-to-boost-financial-services-ai-regulation</link>
                                                                            <description>
                            <![CDATA[ The regulator has called for another review about whether AI needs to be regulated ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">EAy6RbbFJtW4kdgsqXDYjf</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/DMbUEeGtC8v9ZE4MVJneM9-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 08 Jul 2026 08:24:19 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:27 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Policy and Legislation]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/DMbUEeGtC8v9ZE4MVJneM9-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Consumer banking concept image showing rows of green credit cards in a digitized environment.]]></media:description>                                                            <media:text><![CDATA[Consumer banking concept image showing rows of green credit cards in a digitized environment.]]></media:text>
                                <media:title type="plain"><![CDATA[Consumer banking concept image showing rows of green credit cards in a digitized environment.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/DMbUEeGtC8v9ZE4MVJneM9-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Stronger powers could be needed to regulate the use of AI in financial services as consumers flock to the technology for advice, according to the Financial Conduct Authority (FCA).</p><p>Following an investigation into the technology and its impact on consumer finance activities, the regulator has recommended a formal review into how to regulate the use of general-purpose LLMs across the industry. </p><p>The FCA report – known as the Mills Review after its author – noted that companies and consumers are increasingly delegating decisions about finances to AI and AI-powered systems. </p><p>A survey by the regulator found that one-in-five adults were already open to the idea of general-purpose <a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> – such as Claude, <a href="https://www.itpro.com/technology/artificial-intelligence/openai-chatgpt-superapp-overhaul-public-listing">ChatGPT</a>, or Gemini – making financial decisions on their behalf, in particular around debt advice, pensions, and investments. </p><p>While that could help reduce existing challenges such as "advice gaps" and friction around switching services, the FCA warned about risks including fraud and scams, as well as disrupting market structure and competition. </p><p>"AI offers a once-in-a-generation chance to close the information asymmetries and frictions that have long left people making poor financial decisions," said FCA executive director and report author Sheldon Mills in the forward to the review. </p><p>"The right spur, in retail financial services, is ensuring consumers can make healthy ones; regulation, whether supportive or restrictive, should serve that outcome."</p><p>The FCA noted that many consumers may not be aware that they have no formal recourse if something goes wrong, and said some assume they receive equal protection akin to traditional sources of advice. </p><p>Emma Banymandhub, CEO of The Payments Association, agreed that the gap in understanding could cause issues and urged caution on the part of consumers. </p><p>"Consumers may be increasingly comfortable using AI agents for routine tasks such as weekly shopping, but AI-driven savings and investment decisions present a very different set of challenges," Banymandhub commented. </p><p>"While AI can increasingly explain investment concepts and analyse financial information, personalised investment recommendations remain subject to important regulatory constraints."</p><h2 id="ai-gains-traction-in-financial-services">AI gains traction in financial services</h2><p>The financial services sector has rapidly emerged as one of the key growth spaces in terms of the use of AI, with a host of major providers across the UK ramping up adoption of the technology. </p><p>As <em>ITPro </em>previously reported, high street lenders such as <a href="https://www.itpro.com/technology/artificial-intelligence/using-generative-ai-as-a-copilot-is-the-sweet-spot-a-look-at-nationwides-ai-approach">Nationwide</a>, Lloyds Banking Group, and <a href="https://www.itpro.com/business/business-strategy/yorkshire-building-society-touts-customer-service-gains-with-ai-agents">Yorkshire Building Society</a> have all made significant strides on this front over the last six months. </p><p>A <a href="https://publications.parliament.uk/pa/cm5901/cmselect/cmtreasy/684/report.html" target="_blank"><u>parliamentary inquiry</u></a> into the use of AI in financial services, published in January 2026, found that the sector “substantially outpaces” others with regard to AI adoption. </p><p>Indeed, around 75% of UK-based financial services firms are now using the technology, with insurers and large banks among the most aggressive in their pursuit of AI adoption. </p><p>Running parallel to this, the use of AI by consumers is also surging, according to research from Lloyds Banking Group. The firm’s 2025 <a href="https://www.lloydsbankinggroup.com/media/press-releases/2025/lloyds-banking-group-2025/28m-adults-using-ai-to-manage-money.html" target="_blank"><u><em>Consumer Digital Index</em></u></a> found that AI has “rapidly become a financial tool for millions across the UK". </p><p>56% of adults – equivalent to around 28 million people – revealed they’d used AI over the preceding 12 months for financial advice. ChatGPT, for example, was referenced as the most popular platform in this regard, used by six-in-10 consumers. </p><h2 id="what-the-mills-review-recommends">What the Mills Review recommends</h2><p>With this in mind, the Mills Review recommended that the existing "regulatory perimeter" needs to be secured and adapted to take in AI's impact on retail financial services – and that should come via a review within the next six months into how AI is affecting the market. </p><p>"The review should examine how consumers use AI including general purpose LLM tools for personal financial management across savings, investments, pensions, mortgages and debt management, and the implications for competition, innovation and growth," the Mills Review said. </p><p>"It should examine the risks of consumer harm, including how far its usage has or will move along the autonomy spectrum, and any impacts on market integrity and the potential for regulatory arbitrage."</p><p>Based on that review's guidance, the FCA could tweak regulation as necessary, the report said. Beyond that, the review called for the FCA to monitor for evidence of harm and new consumer models, engage with providers for better insight into changes, and be ready with intervention tools and mechanisms as necessary. </p><h2 id="frontier-model-monitoring-in-finance">Frontier model monitoring in finance</h2><p>In the longer term, the Mills Review called for the FCA to keep an eye on frontier model capabilities and AI adoption, considering its impact on its regulatory work, in particular where use of AI falls outside of existing protections and once AI agents start taking more action on behalf of consumers. </p><p>The review also called for stronger powers for the FCA so it can look at these issues more widely. </p><p>"As is clear in the report, we need to keep pace with a rapidly changing environment and the principles-based, outcomes focussed approach we’ve taken on AI – relying on the Consumer Duty and Senior Managers Regime – has been critical to us doing so," said Ashley Alder, chair of the FCA. </p><p>"The recommendations build on work the FCA has been doing – not least allowing firms to test their use of AI with us – and our own use of AI to be a smarter regulator, more efficient and effective."</p><p>Banymandhub<strong> </strong>added that walking that balance between enabling AI and protecting consumers and companies was key.  </p><p>"The FCA’s Mills Review reinforces that firms should treat agentic AI as an accountability and governance issue now, while providing greater confidence to innovate responsibly as AI adoption accelerates," she said. </p><p>"AI has enormous potential for financial services, but realising that potential will depend on strong governance, clear accountability and maintaining consumer trust."</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Why the US imposed export controls on Anthropic’s Fable and Mythos models ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Anthropic's Fable 5 and Mythos 5 are back after a three-weeks-long restriction <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-suspends-fabel-and-mythos-systems-for-all-users-after-us-government-claims-jailbreak-risk">saw the security models effectively banned</a>. </p><p>Earlier this month, the US government applied export controls to <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">Claude Mythos</a> and Claude Fable 5, banning any foreign users, inside or outside the US, from either model. </p><p>As Anthropic had no way of checking the nationality of its users, both models were de facto banned from use. Last week, the restriction was loosened to allow Mythos to be used by US organizations deemed trustworthy. </p><p>Now, Anthropic has <a href="https://www.anthropic.com/news/redeploying-fable-5" target="_blank"><u>said</u></a> the restrictions have been mostly dropped following the introduction of new safeguards. </p><p>"Anthropic has agreed to proactively detect and address security risks associated with the models," US Commerce Secretary Howard Lutnick wrote in a letter to the tech company, according to the <a href="https://www.bbc.co.uk/news/articles/cdr42623e1do" target="_blank"><u><em>BBC</em></u></a>, </p><p>Lutnick later added on <a href="https://x.com/howardlutnick/status/2072100729603452965" target="_blank"><u>social media</u></a>: "Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI."</p><p>Anthropic said it was still working to expand access to Mythos 5 to a wider set of partners in its <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-lets-glasswing-partners-publicly-share-mythos-flaws">Project Glasswing</a> program – which gives access to the security model in a managed way – and hoped that would include domestic and international users. </p><p>Fable 5 should now be available again for all Claude users, with access reenabled for AWS, Google Cloud, and Microsoft Foundry as quickly as possible.</p><h2 id="why-mythos-and-fable-were-restricted">Why Mythos and Fable were restricted</h2><p>Fable 5 and Mythos 5 are both security focused models, sharing the same underlying model, Anthropic notes. </p><p>As <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-just-launched-claude-fable-5-its-first-mythos-class-ai-model-but-it-has-new-safeguards-to-prevent-misuse-and-will-fall-back-to-opus-4-8-for-high-risk-queries"><em>ITPro </em>reported in early June</a>, Fable 5 was built with "strong safeguards to make it safe for general use". </p><p>Mythos, meanwhile, has fewer protections in place, and as such was released to a limited number of trusted partners under Project Glasswing for the express purpose of defensive security.</p><p>Just days after the release of Fable 5, the government applied strict export controls. That was driven by Amazon researchers finding a way to dodge Fable 5's safeguards to use it to hunt for flaws in software and produce exploits. </p><p>Anthropic argued that most of its models could actually uncover the same vulnerability and exploit demonstration, and that the Amazon research didn't make use of any "unique Mythos-level cyber capabilities". </p><p>The workaround discovered by Amazon has now been blocked for most instances, with any users attempting similar techniques downgraded to a lesser model. Anthropic admitted that more "benign requests" may be flagged accidentally now, though it hopes to reduce such false positives as the system is improved. </p><p>"One particularly important safety mechanism involves classifiers — smaller automated AI systems that, during an interaction, detect when the model is asked to perform a potentially harmful cybersecurity task (or produces potentially harmful outputs)," the company said. </p><p>"When this occurs, the classifiers block the model from responding to requests. The ultimate goal of these classifiers is to prevent the model from engaging in uniquely dangerous behaviors."</p><p>Beyond the new safety classifiers and other safeguards, Anthropic is working with industry partners including Amazon, Microsoft, and Google to develop a framework for addressing <a href="https://www.itpro.com/technology/artificial-intelligence/some-of-the-most-popular-open-weight-ai-models-show-profound-susceptibility-to-jailbreak-techniques">AI model ‘jailbreaking’ techniques</a>, developing a system for judging the severity of such attacks. </p><p>Plus, Anthropic said it would work more closely with the US government, offering expanded early access for models that could impact national security and sharing information about jailbreaks or misuse patterns. </p><p>The company called for the same rules to apply to its rivals: "These rules should be codified in strong regulation and applied equally across frontier model developers."</p><h2 id="mixed-industry-reactions">Mixed industry reactions</h2><p>It's no surprise this was sorted quickly. Keven Knight, CEO, of <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>firm, Talion, noted that the US government would not have wanted to risk slowing down <a href="https://www.itpro.com/technology/artificial-intelligence/what-would-pausing-ai-development-actually-achieve">AI development</a>, particularly in security. </p><p>"Given that China is steadily advancing with its advanced AI model, with a platform that mirrors Mythos being launched by 360 Security Technology last week, the US could not afford to limit access to Anthropic's platforms for much longer," Knight said. </p><p>"Restricting access to the platforms would only leave western organisations on the backfoot, and at a time when the AI arms race is really heating up, this would be dangerous."</p><p>Though the export ban is now loosened, Andrew Bloster, senior R&amp;D manager at Black Duck, said that won't end the "chilling effect" the incident has had on the use of such models. </p><p>"Security leaders globally are now wary to depend on AI as a service models that can be launched with much fanfare and then pulled from the global market," he said, pointing to the rise of sovereign AI. </p><p>Bloster added that "technology leaders are more concerned about resilience, data security, and cost, and Fable being permitted back on the world stage might not be enough to earn back global trust."</p><p>Though Anthropic was the first company to jump through these hoops, it actually tried to find a good balance between safety and use via Glasswing, said Mike Britton, CIO at Abnormal AI. </p><p>"On Mythos, the controlled rollout to a vetted set of US organizations is actually the right model. Limiting access to trusted institutions, with use cases focused on finding and remediating vulnerabilities, is how you derive real security value from a model at this capability level without creating new risks," said Britton. </p><p>"The goal should be more secure software — and that's achievable if the access controls stay disciplined over time."</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/why-the-us-imposed-export-controls-on-anthropics-fable-and-mythos-models-and-why-theyve-been-lifted</link>
                                                                            <description>
                            <![CDATA[ Anthropic tightens up safeguards and offers expanded early access to US government to end export control issues ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">vvdbbN3TAkaAqxB6e2PzRf</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/7q4RMcThiHYVFNwwwNrEv8-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 02 Jul 2026 10:53:04 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:10 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/7q4RMcThiHYVFNwwwNrEv8-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Dario Amodei, CEO and co-founder of Anthropic, pictured during an interview on &quot;The Circuit with Emily Chang&quot; at the company&#039;s headquarters in San Francisco, USA.]]></media:description>                                                            <media:text><![CDATA[Dario Amodei, CEO and co-founder of Anthropic, pictured during an interview on &quot;The Circuit with Emily Chang&quot; at the company&#039;s headquarters in San Francisco, USA.]]></media:text>
                                <media:title type="plain"><![CDATA[Dario Amodei, CEO and co-founder of Anthropic, pictured during an interview on &quot;The Circuit with Emily Chang&quot; at the company&#039;s headquarters in San Francisco, USA.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/7q4RMcThiHYVFNwwwNrEv8-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Anthropic's Fable 5 and Mythos 5 are back after a three-weeks-long restriction <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-suspends-fabel-and-mythos-systems-for-all-users-after-us-government-claims-jailbreak-risk">saw the security models effectively banned</a>. </p><p>Earlier this month, the US government applied export controls to <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">Claude Mythos</a> and Claude Fable 5, banning any foreign users, inside or outside the US, from either model. </p><p>As Anthropic had no way of checking the nationality of its users, both models were de facto banned from use. Last week, the restriction was loosened to allow Mythos to be used by US organizations deemed trustworthy. </p><p>Now, Anthropic has <a href="https://www.anthropic.com/news/redeploying-fable-5" target="_blank"><u>said</u></a> the restrictions have been mostly dropped following the introduction of new safeguards. </p><p>"Anthropic has agreed to proactively detect and address security risks associated with the models," US Commerce Secretary Howard Lutnick wrote in a letter to the tech company, according to the <a href="https://www.bbc.co.uk/news/articles/cdr42623e1do" target="_blank"><u><em>BBC</em></u></a>, </p><p>Lutnick later added on <a href="https://x.com/howardlutnick/status/2072100729603452965" target="_blank"><u>social media</u></a>: "Over the past two weeks, we have worked closely with Anthropic to analyze and approve Fable 5 to ensure alignment across the US Government and strengthen America’s leadership in AI."</p><p>Anthropic said it was still working to expand access to Mythos 5 to a wider set of partners in its <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-lets-glasswing-partners-publicly-share-mythos-flaws">Project Glasswing</a> program – which gives access to the security model in a managed way – and hoped that would include domestic and international users. </p><p>Fable 5 should now be available again for all Claude users, with access reenabled for AWS, Google Cloud, and Microsoft Foundry as quickly as possible.</p><h2 id="why-mythos-and-fable-were-restricted">Why Mythos and Fable were restricted</h2><p>Fable 5 and Mythos 5 are both security focused models, sharing the same underlying model, Anthropic notes. </p><p>As <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-just-launched-claude-fable-5-its-first-mythos-class-ai-model-but-it-has-new-safeguards-to-prevent-misuse-and-will-fall-back-to-opus-4-8-for-high-risk-queries"><em>ITPro </em>reported in early June</a>, Fable 5 was built with "strong safeguards to make it safe for general use". </p><p>Mythos, meanwhile, has fewer protections in place, and as such was released to a limited number of trusted partners under Project Glasswing for the express purpose of defensive security.</p><p>Just days after the release of Fable 5, the government applied strict export controls. That was driven by Amazon researchers finding a way to dodge Fable 5's safeguards to use it to hunt for flaws in software and produce exploits. </p><p>Anthropic argued that most of its models could actually uncover the same vulnerability and exploit demonstration, and that the Amazon research didn't make use of any "unique Mythos-level cyber capabilities". </p><p>The workaround discovered by Amazon has now been blocked for most instances, with any users attempting similar techniques downgraded to a lesser model. Anthropic admitted that more "benign requests" may be flagged accidentally now, though it hopes to reduce such false positives as the system is improved. </p><p>"One particularly important safety mechanism involves classifiers — smaller automated AI systems that, during an interaction, detect when the model is asked to perform a potentially harmful cybersecurity task (or produces potentially harmful outputs)," the company said. </p><p>"When this occurs, the classifiers block the model from responding to requests. The ultimate goal of these classifiers is to prevent the model from engaging in uniquely dangerous behaviors."</p><p>Beyond the new safety classifiers and other safeguards, Anthropic is working with industry partners including Amazon, Microsoft, and Google to develop a framework for addressing <a href="https://www.itpro.com/technology/artificial-intelligence/some-of-the-most-popular-open-weight-ai-models-show-profound-susceptibility-to-jailbreak-techniques">AI model ‘jailbreaking’ techniques</a>, developing a system for judging the severity of such attacks. </p><p>Plus, Anthropic said it would work more closely with the US government, offering expanded early access for models that could impact national security and sharing information about jailbreaks or misuse patterns. </p><p>The company called for the same rules to apply to its rivals: "These rules should be codified in strong regulation and applied equally across frontier model developers."</p><h2 id="mixed-industry-reactions">Mixed industry reactions</h2><p>It's no surprise this was sorted quickly. Keven Knight, CEO, of <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>firm, Talion, noted that the US government would not have wanted to risk slowing down <a href="https://www.itpro.com/technology/artificial-intelligence/what-would-pausing-ai-development-actually-achieve">AI development</a>, particularly in security. </p><p>"Given that China is steadily advancing with its advanced AI model, with a platform that mirrors Mythos being launched by 360 Security Technology last week, the US could not afford to limit access to Anthropic's platforms for much longer," Knight said. </p><p>"Restricting access to the platforms would only leave western organisations on the backfoot, and at a time when the AI arms race is really heating up, this would be dangerous."</p><p>Though the export ban is now loosened, Andrew Bloster, senior R&amp;D manager at Black Duck, said that won't end the "chilling effect" the incident has had on the use of such models. </p><p>"Security leaders globally are now wary to depend on AI as a service models that can be launched with much fanfare and then pulled from the global market," he said, pointing to the rise of sovereign AI. </p><p>Bloster added that "technology leaders are more concerned about resilience, data security, and cost, and Fable being permitted back on the world stage might not be enough to earn back global trust."</p><p>Though Anthropic was the first company to jump through these hoops, it actually tried to find a good balance between safety and use via Glasswing, said Mike Britton, CIO at Abnormal AI. </p><p>"On Mythos, the controlled rollout to a vetted set of US organizations is actually the right model. Limiting access to trusted institutions, with use cases focused on finding and remediating vulnerabilities, is how you derive real security value from a model at this capability level without creating new risks," said Britton. </p><p>"The goal should be more secure software — and that's achievable if the access controls stay disciplined over time."</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Three quarters of firms have halted AI projects over safety and security concerns ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Three quarters of companies have had to halt AI projects because of safety and security concerns over the last year, and systems like Claude Mythos are only going to make security teams' jobs harder. </p><p>That's according to a pair of surveys by Aikido and CybaVerse, which highlighted that AI is clearly becoming a security problem – a fact echoed by <a href="https://www.itpro.com/security/five-eyes-agencies-sound-alarm-over-risky-agentic-ai-deployments">Five Eyes</a> cybersecurity agencies warning that leaders need to act now to <a href="https://www.ncsc.gov.uk/news/the-ai-shift-in-cyber-risk-why-leaders-must-act-now" target="_blank">stay ahead of AI-related security risks</a>.</p><p>"Adversaries are already using AI to move faster and more effectively. Defenders must do the same," said the UK's <a href="https://www.itpro.com/security/what-is-the-national-cyber-security-centre-ncsc-and-what-does-it-do">National Cyber Security Centre</a> in a <a href="https://www.ncsc.gov.uk/news/the-ai-shift-in-cyber-risk-why-leaders-must-act-now" target="_blank"><u>blog post</u></a>. </p><p>Rolling out AI to solve the challenges of AI may not prove a simple solution, however. A survey from Aikido Security found that 76% of organizations polled had to stop, restrict, or <a href="https://www.itpro.com/technology/artificial-intelligence/the-ai-rollback-nobody-wants-to-talk-about">roll back AI projects</a> over the last year. </p><p>That figure rises to 98% for more active teams that are shipping multiple times a day. </p><p>The report found that seven-in-ten companies had a security issue that was harder to detect, investigate, or remediate because of AI, which again rose to 86% for daily shippers.</p><p>According to Aikido, the key challenge here lies in velocity. AI has accelerated the pace of development, meaning security teams have less time to spot flaws — and more arrive by the time the first round can be fixed.</p><p>Notably, the survey found that three quarters of those polled are rolling out significant production changes weekly – or faster – but only two-in-ten are validating security at that rate. </p><p>Because of that, nearly eight-in-ten are concerned about vulnerabilities being introduced between tests and half say test findings are outdated by the time they see them. </p><h2 id="rising-ai-security-concerns">Rising AI security concerns</h2><p>It's perhaps no surprise then that nine-in-ten security professionals see the rise of security-focused models such as <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">Anthropic's Claude Mythos</a> as increasing cyber risk for companies</p><p>According to a survey by CybaVerse, 86% of polled cyber pros believe AI systems like Mythos will cut the time it takes for hackers to spot and exploit flaws, leading to more patching. More than two-thirds said their employer lacks the budget to address that increased workload. </p><p>“Advanced AI platforms were blasted into the public domain and organizations had no time to prepare for the impact they would have on their cyber defences," said Oliver Spence, CEO of CybaVerse. </p><p>"Now that some of the largest technology companies in the world have access to these platforms, we are already seeing an increase in the volume of vulnerabilities being identified and disclosed, with the latest Patch Tuesday being the largest on record."</p><p>Looking ahead, security professionals aren't optimistic that AI will make the situation better. Indeed, three quarters believe an advanced AI system will eventually be weaponized by cyber criminals. </p><p>"This is something organizations must be prepared for because bulletproof security doesn’t exist," said Spence. </p><h2 id="back-to-basics">Back to basics</h2><p>Spence added that core defense techniques haven't changed, even if AI has accelerated the <a href="https://www.itpro.com/security/brace-yourselves-for-a-vulnerability-explosion-forescout-warns">pace of vulnerability discovery</a>. </p><p>"Security teams still need visibility of their assets, they still need strong vulnerability management processes and they still need to prioritise remediation efforts based on risk," he said. </p><p>"Advanced AI may change the speed and scale of cyber threats, but the organizations that maintain strong <a href="https://www.itpro.com/security/cyber-requirements-stress-perspective">cyber hygiene</a> and focus on reducing exposure to their most significant risks will always be in the strongest position to defend themselves,” added Spence.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/three-quarters-of-firms-have-halted-ai-projects-over-safety-and-security-concerns-and-cyber-pros-think-things-will-deteriorate-as-models-like-claude-mythos-improve</link>
                                                                            <description>
                            <![CDATA[ AI has become a leading problem for enterprise security teams, they can't automate their way out of trouble ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">7PbH4iWL5JStoNxAD8jncF</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/A2eUMwBBjVbDZpzbyz9BrQ-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 24 Jun 2026 07:35:51 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:14 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/A2eUMwBBjVbDZpzbyz9BrQ-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[AI security concept image showing a digitized padlock symbol with &#039;AI&#039; symbol, connected to data points with multi-colored data flows emanating from each point.]]></media:description>                                                            <media:text><![CDATA[AI security concept image showing a digitized padlock symbol with &#039;AI&#039; symbol, connected to data points with multi-colored data flows emanating from each point.]]></media:text>
                                <media:title type="plain"><![CDATA[AI security concept image showing a digitized padlock symbol with &#039;AI&#039; symbol, connected to data points with multi-colored data flows emanating from each point.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/A2eUMwBBjVbDZpzbyz9BrQ-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Three quarters of companies have had to halt AI projects because of safety and security concerns over the last year, and systems like Claude Mythos are only going to make security teams' jobs harder. </p><p>That's according to a pair of surveys by Aikido and CybaVerse, which highlighted that AI is clearly becoming a security problem – a fact echoed by <a href="https://www.itpro.com/security/five-eyes-agencies-sound-alarm-over-risky-agentic-ai-deployments">Five Eyes</a> cybersecurity agencies warning that leaders need to act now to <a href="https://www.ncsc.gov.uk/news/the-ai-shift-in-cyber-risk-why-leaders-must-act-now" target="_blank">stay ahead of AI-related security risks</a>.</p><p>"Adversaries are already using AI to move faster and more effectively. Defenders must do the same," said the UK's <a href="https://www.itpro.com/security/what-is-the-national-cyber-security-centre-ncsc-and-what-does-it-do">National Cyber Security Centre</a> in a <a href="https://www.ncsc.gov.uk/news/the-ai-shift-in-cyber-risk-why-leaders-must-act-now" target="_blank"><u>blog post</u></a>. </p><p>Rolling out AI to solve the challenges of AI may not prove a simple solution, however. A survey from Aikido Security found that 76% of organizations polled had to stop, restrict, or <a href="https://www.itpro.com/technology/artificial-intelligence/the-ai-rollback-nobody-wants-to-talk-about">roll back AI projects</a> over the last year. </p><p>That figure rises to 98% for more active teams that are shipping multiple times a day. </p><p>The report found that seven-in-ten companies had a security issue that was harder to detect, investigate, or remediate because of AI, which again rose to 86% for daily shippers.</p><p>According to Aikido, the key challenge here lies in velocity. AI has accelerated the pace of development, meaning security teams have less time to spot flaws — and more arrive by the time the first round can be fixed.</p><p>Notably, the survey found that three quarters of those polled are rolling out significant production changes weekly – or faster – but only two-in-ten are validating security at that rate. </p><p>Because of that, nearly eight-in-ten are concerned about vulnerabilities being introduced between tests and half say test findings are outdated by the time they see them. </p><h2 id="rising-ai-security-concerns">Rising AI security concerns</h2><p>It's perhaps no surprise then that nine-in-ten security professionals see the rise of security-focused models such as <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">Anthropic's Claude Mythos</a> as increasing cyber risk for companies</p><p>According to a survey by CybaVerse, 86% of polled cyber pros believe AI systems like Mythos will cut the time it takes for hackers to spot and exploit flaws, leading to more patching. More than two-thirds said their employer lacks the budget to address that increased workload. </p><p>“Advanced AI platforms were blasted into the public domain and organizations had no time to prepare for the impact they would have on their cyber defences," said Oliver Spence, CEO of CybaVerse. </p><p>"Now that some of the largest technology companies in the world have access to these platforms, we are already seeing an increase in the volume of vulnerabilities being identified and disclosed, with the latest Patch Tuesday being the largest on record."</p><p>Looking ahead, security professionals aren't optimistic that AI will make the situation better. Indeed, three quarters believe an advanced AI system will eventually be weaponized by cyber criminals. </p><p>"This is something organizations must be prepared for because bulletproof security doesn’t exist," said Spence. </p><h2 id="back-to-basics">Back to basics</h2><p>Spence added that core defense techniques haven't changed, even if AI has accelerated the <a href="https://www.itpro.com/security/brace-yourselves-for-a-vulnerability-explosion-forescout-warns">pace of vulnerability discovery</a>. </p><p>"Security teams still need visibility of their assets, they still need strong vulnerability management processes and they still need to prioritise remediation efforts based on risk," he said. </p><p>"Advanced AI may change the speed and scale of cyber threats, but the organizations that maintain strong <a href="https://www.itpro.com/security/cyber-requirements-stress-perspective">cyber hygiene</a> and focus on reducing exposure to their most significant risks will always be in the strongest position to defend themselves,” added Spence.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Workday hits back amid lawsuit claiming AI recruitment discrimination ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Three years ago, a Californian man claimed that he'd been rejected hundreds of times by Workday's AI-driven HR system – now a judge has ruled that a class action based on the complaint will be heard in court.</p><p>US District Judge Rita Lin has <a href="https://www.govinfo.gov/content/pkg/USCOURTS-cand-3_23-cv-00770/pdf/USCOURTS-cand-3_23-cv-00770-13.pdf" target="_blank"><u>ruled</u></a> that Workday must face the claims brought against it, which includes claims that the company's AI-based candidate screening system discriminated on the grounds of age, race, and other protected characteristics. </p><p>With the case set to proceed, this makes it the first to consider the impact of letting AI-powered software make decisions on job candidates – a significant case given the rise of recruitment. </p><p>Indeed, a <a href="https://sqmagazine.co.uk/ai-recruitment-statistics/" target="_blank"><u>report from HireVue suggests</u></a> seven-in-ten companies are using AI in HR and recruitment processes. The shift to AI hiring isn't popular, either, with two-thirds of Americans telling Pew that they didn't want to apply for a job via AI.</p><p>A spokesperson for Workday told <em>ITPro </em>the “claims in the suit are false” and dismissed claims that AI recruiting tools make hiring decisions. </p><p>“Workday’s AI recruiting tools don’t make hiring decisions in California or anywhere else,” the spokesperson said. “Our customers maintain full control of their hiring process and our tools are designed with human oversight at their core.”</p><p>Workday added that its technology “looks only at job qualifications, not protected traits like race, age, or disability”. </p><p>“We rigorously test our products as part of our Responsible AI program to confirm our tools do not harm protected groups.”</p><h2 id="workday-case-considerations">Workday case considerations</h2><p>Lin ruled that anti-discrimination laws do indeed apply when Workday's software is used outside of California, where the company is based and where the claim was filed. </p><p>Workday's lawyers had argued that "it makes no sense for a Texas employer with a Texas applicant who will perform work in Texas," but Lin disagreed. </p><p>Lin also refused to drop a claim alleging that the software filters out candidates using "proxy indicators", including gaps in employment history, which could impact those with disabilities or illnesses. </p><p>One claim was dismissed, however. This sought to include discrimination against Asian Americans in the suit, on the grounds that proper procedure wasn't followed. </p><h2 id="history-of-the-case">History of the case</h2><p>In 2023, Derek Mobley first <a href="https://www.itpro.com/business/policy-legislation/370133/workday-hit-with-claims-its-ai-hiring-systems-are-discriminatory"><u>sued Workday claiming</u></a> that he had been rejected by the company's software, which is used by other companies to sift through candidates, between 80 and 100 times. </p><p>Mobley believed that the Workday pre-selection system was discriminating against him because he is a black American, over 40 and suffers disabilities. At the time, Workday said the lawsuit was "without merit."</p><p>Workday attempted to have the case thrown out in 2024, but Judge Lin dismissed those challenges, ruling that the company wasn't an employment agency but could for the case be considered an employer. </p><p>She dismissed claims that the discrimination was intentional. At the time, Workday <a href="https://www.reuters.com/legal/litigation/workday-must-face-novel-bias-lawsuit-over-ai-screening-software-2024-07-15/" target="_blank"><u>said</u></a> it was confident that the remaining allegations would be "easily refuted." </p><p>Earlier this year, Mobley was joined by four other would-be candidates, with the judge ruling they could be considered together as a class action. </p><p>One of the new plaintiffs, Jill Hughes, <a href="https://www.itpro.com/business/careers-and-training/workday-faces-lawsuit-over-alleged-ai-bias"><u>said she submitted hundreds of applications</u></a> that were rejected, with some responses saying she didn't meet the job's minimum requirement, when she actually did. </p><h2 id="the-ai-recruitment-conundrum">The AI recruitment conundrum</h2><p>With the case set to go ahead, it could reveal much sought after details about how AI is being used in recruitment and how it influences who is being hired – and not. </p><p>The case comes as AI has overrun recruitment, with <a href="https://www.itpro.com/business/careers-and-training/ai-resume-screening-recruiter-chatbots-and-ghost-jobs-are-causing-havoc-for-struggling-entry-level-workers"><u>companies using it to power everything</u></a> from sifting through CVs to conducting interviews, while <a href="https://www.itpro.com/business/careers-and-training/uk-jobseekers-could-be-using-ai-to-beef-up-cvs-lie-on-applications-and-complete-skills-tests-heres-why-you-really-shouldnt-do-that"><u>would-be candidates use AI</u></a> to find open roles, apply automatically, and boost their responses in interviews.</p><p>This has created a situation where AI is essentially talking to AI in order to hire a human for a job. <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-job-applications-ai"><u>Anthropic raised eyebrows</u></a> last year when it told candidates not to use AI when applying, for example. </p><p>Workday isn't the first company to be caught out by the risks of using AI for recruitment. Back in 2018, <a href="https://www.itpro.com/machine-learning/32083/ai-recruitment-tool-pulled-by-amazon-for-sex-bias"><u>Amazon was forced to pull an AI tool</u></a> it was testing for recruitment after it became clear the system was methodically filtering out CVs submitted by women. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/the-claims-in-the-suit-are-false-workday-hits-back-amid-lawsuit-claiming-ai-recruitment-discrimination</link>
                                                                            <description>
                            <![CDATA[ Is AI hiring discriminatory? A California judge has given the go-ahead for a class action suit against Workday ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">DxMdgsty66ZJTGa9egKGqT</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/K3AxvFCdwkHbGaziWyhvKj-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 23 Jun 2026 11:30:37 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:22 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/K3AxvFCdwkHbGaziWyhvKj-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Workday logo and branding pictured on a building facade with blue background.]]></media:description>                                                            <media:text><![CDATA[Workday logo and branding pictured on a building facade with blue background.]]></media:text>
                                <media:title type="plain"><![CDATA[Workday logo and branding pictured on a building facade with blue background.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/K3AxvFCdwkHbGaziWyhvKj-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Three years ago, a Californian man claimed that he'd been rejected hundreds of times by Workday's AI-driven HR system – now a judge has ruled that a class action based on the complaint will be heard in court.</p><p>US District Judge Rita Lin has <a href="https://www.govinfo.gov/content/pkg/USCOURTS-cand-3_23-cv-00770/pdf/USCOURTS-cand-3_23-cv-00770-13.pdf" target="_blank"><u>ruled</u></a> that Workday must face the claims brought against it, which includes claims that the company's AI-based candidate screening system discriminated on the grounds of age, race, and other protected characteristics. </p><p>With the case set to proceed, this makes it the first to consider the impact of letting AI-powered software make decisions on job candidates – a significant case given the rise of recruitment. </p><p>Indeed, a <a href="https://sqmagazine.co.uk/ai-recruitment-statistics/" target="_blank"><u>report from HireVue suggests</u></a> seven-in-ten companies are using AI in HR and recruitment processes. The shift to AI hiring isn't popular, either, with two-thirds of Americans telling Pew that they didn't want to apply for a job via AI.</p><p>A spokesperson for Workday told <em>ITPro </em>the “claims in the suit are false” and dismissed claims that AI recruiting tools make hiring decisions. </p><p>“Workday’s AI recruiting tools don’t make hiring decisions in California or anywhere else,” the spokesperson said. “Our customers maintain full control of their hiring process and our tools are designed with human oversight at their core.”</p><p>Workday added that its technology “looks only at job qualifications, not protected traits like race, age, or disability”. </p><p>“We rigorously test our products as part of our Responsible AI program to confirm our tools do not harm protected groups.”</p><h2 id="workday-case-considerations">Workday case considerations</h2><p>Lin ruled that anti-discrimination laws do indeed apply when Workday's software is used outside of California, where the company is based and where the claim was filed. </p><p>Workday's lawyers had argued that "it makes no sense for a Texas employer with a Texas applicant who will perform work in Texas," but Lin disagreed. </p><p>Lin also refused to drop a claim alleging that the software filters out candidates using "proxy indicators", including gaps in employment history, which could impact those with disabilities or illnesses. </p><p>One claim was dismissed, however. This sought to include discrimination against Asian Americans in the suit, on the grounds that proper procedure wasn't followed. </p><h2 id="history-of-the-case">History of the case</h2><p>In 2023, Derek Mobley first <a href="https://www.itpro.com/business/policy-legislation/370133/workday-hit-with-claims-its-ai-hiring-systems-are-discriminatory"><u>sued Workday claiming</u></a> that he had been rejected by the company's software, which is used by other companies to sift through candidates, between 80 and 100 times. </p><p>Mobley believed that the Workday pre-selection system was discriminating against him because he is a black American, over 40 and suffers disabilities. At the time, Workday said the lawsuit was "without merit."</p><p>Workday attempted to have the case thrown out in 2024, but Judge Lin dismissed those challenges, ruling that the company wasn't an employment agency but could for the case be considered an employer. </p><p>She dismissed claims that the discrimination was intentional. At the time, Workday <a href="https://www.reuters.com/legal/litigation/workday-must-face-novel-bias-lawsuit-over-ai-screening-software-2024-07-15/" target="_blank"><u>said</u></a> it was confident that the remaining allegations would be "easily refuted." </p><p>Earlier this year, Mobley was joined by four other would-be candidates, with the judge ruling they could be considered together as a class action. </p><p>One of the new plaintiffs, Jill Hughes, <a href="https://www.itpro.com/business/careers-and-training/workday-faces-lawsuit-over-alleged-ai-bias"><u>said she submitted hundreds of applications</u></a> that were rejected, with some responses saying she didn't meet the job's minimum requirement, when she actually did. </p><h2 id="the-ai-recruitment-conundrum">The AI recruitment conundrum</h2><p>With the case set to go ahead, it could reveal much sought after details about how AI is being used in recruitment and how it influences who is being hired – and not. </p><p>The case comes as AI has overrun recruitment, with <a href="https://www.itpro.com/business/careers-and-training/ai-resume-screening-recruiter-chatbots-and-ghost-jobs-are-causing-havoc-for-struggling-entry-level-workers"><u>companies using it to power everything</u></a> from sifting through CVs to conducting interviews, while <a href="https://www.itpro.com/business/careers-and-training/uk-jobseekers-could-be-using-ai-to-beef-up-cvs-lie-on-applications-and-complete-skills-tests-heres-why-you-really-shouldnt-do-that"><u>would-be candidates use AI</u></a> to find open roles, apply automatically, and boost their responses in interviews.</p><p>This has created a situation where AI is essentially talking to AI in order to hire a human for a job. <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-job-applications-ai"><u>Anthropic raised eyebrows</u></a> last year when it told candidates not to use AI when applying, for example. </p><p>Workday isn't the first company to be caught out by the risks of using AI for recruitment. Back in 2018, <a href="https://www.itpro.com/machine-learning/32083/ai-recruitment-tool-pulled-by-amazon-for-sex-bias"><u>Amazon was forced to pull an AI tool</u></a> it was testing for recruitment after it became clear the system was methodically filtering out CVs submitted by women. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ IT teams are bullish on AI tools, but they’re worried security practices can’t keep pace ]]></title>
                                                                                                <dc:content><![CDATA[ <p>IT teams are growing increasingly concerned about <a href="https://www.itpro.com/security/the-key-risks-security-teams-face-in-2026">AI-related risks</a> amidst continued adoption, according to new research. </p><p>Findings from Heimdal’s <em>State of AI Risk Management in 2026</em> report show <a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> are now commonplace across most IT estates, with teams often running several at one. </p><p><a href="https://www.itpro.com/technology/artificial-intelligence-ai/369965/what-is-chatgpt-and-what-does-it-mean-for-businesses">ChatGPT</a>, for example, runs in nearly three-quarters (71%) of UK IT environments, while <a href="https://www.itpro.com/technology/artificial-intelligence/microsoft-copilot-review-ai-baked-into-your-apps">Microsoft Copilot</a> is present in 68%. These same figures are reflected across the Atlantic, with US-based IT teams often using a combination of multiple different AI solutions. </p><p>AI tools are playing a key role in reducing workloads, the study noted, which IT teams highlighted as the most positive benefits of the technology. </p><p>Nearly three-quarters of IT and security teams said they lose around a quarter of their week to “repetitive, low-value work” - and AI is helping reduce that manual toil. </p><p>Indeed, teams facing the highest levels of operational load are often ranked among the most optimistic when it comes to AI. More than half (59%) of US teams said they expect AI to alleviate pressure, while 55% in the UK expect the same. </p><h2 id="ai-related-risks-are-haunting-it-teams">AI-related risks are haunting IT teams</h2><p>Despite this optimism, a key recurring concern among IT leaders is that controls and security capabilities haven’t kept pace with the rate of adoption. Heimdal noted that only four-in-ten teams rate their security stack as “ready for AI-related risk”. </p><p>Teams are increasingly concerned about data leakage, for example, with 56% of UK respondents highlighting this risk. Visibility is a major issue in this regard, Heimdal found. </p><p>UK teams with full <a href="https://www.itpro.com/security/it-leaders-are-facing-major-work-device-blind-spots-and-its-putting-security-at-risk">visibility into AI use</a> were most likely to flag data leakage as a leading concern, compared to just 27% of those with no visibility. In the US, 59% of teams with full visibility also highlighted data leakage as a key concern. </p><p>While concerns over potential risks typically mount as AI tools are integrated, Heimdal noted that unauthorized AI use, or ‘shadow AI’, is also a recurring problem for enterprises. </p><p>Heimdal specifically highlighted the <a href="https://www.itpro.com/security/cyber-attacks/the-salesloft-hackers-claim-they-have-1-5-billion-compromised-salesforce-records">Salesloft Drift breach in August 2025</a> as a key example of how poor AI-related visibility can impact organisations. </p><p>The incident saw threat actors steal OAuth tokens for Drift’s AI chatbot integration with Salesforce, using these to extract data from several hundred Salesforce instances. </p><p>A host of organisations, including Cloudflare, Palo Alto Networks, and Zscaler were impacted in the attacks.</p><p>“Drift was the AI tool. Salesforce held the data,” the company noted in a <a href="https://heimdalsecurity.com/blog/state-ai-risk-management/" target="_blank"><u>blog post</u></a>.</p><p>“Most of the affected teams had never personally provisioned Drift,” it added. “A third-party AI chatbot, plugged in through an OAuth grant few had recently reviewed, became the way in.”</p><h2 id="contrasting-priorities">Contrasting priorities</h2><p>Perception of AI-related risk among frontline practitioners and executives is a major problem, according to Heimdal. Indeed, “executive confidence” in AI security is a repeated point of friction when it comes to governance and risk management. </p><p>In the US, for example, 29% of executives said AI risk is under control, yet just 7% of practitioners agreed. In the UK, meanwhile, these figures stand at 18% against 11%. </p><p>Adam Pilton, cybersecurity advisor at Heimdal, said this shows many organizations still aren’t fully aligned on how they manage AI risk. </p><p>"Misplaced confidence is one of the most dangerous things in security. This data shows executives are far more confident that AI risk is under control than the evidence supports. Most of the conversation right now is about productivity, when the bigger question is how AI can be turned against the business,” he said. </p><p>“The report shows the gap between how secure leaders feel and how secure they actually are.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/it-teams-are-bullish-on-ai-tools-but-theyre-worried-security-practices-cant-keep-pace</link>
                                                                            <description>
                            <![CDATA[ Executives and IT teams are at odds over the risks associated with AI adoption ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">MccjwugkSse6xqBn5vzkeG</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/4CeYDBXGHtWfkgtgjJ2dUi-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Mon, 22 Jun 2026 14:33:47 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:13 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/4CeYDBXGHtWfkgtgjJ2dUi-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[A wide angle shot of IT workers in an office, lit by large modern windows showing a business park behind them.]]></media:description>                                                            <media:text><![CDATA[A wide angle shot of IT workers in an office, lit by large modern windows showing a business park behind them.]]></media:text>
                                <media:title type="plain"><![CDATA[A wide angle shot of IT workers in an office, lit by large modern windows showing a business park behind them.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/4CeYDBXGHtWfkgtgjJ2dUi-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>IT teams are growing increasingly concerned about <a href="https://www.itpro.com/security/the-key-risks-security-teams-face-in-2026">AI-related risks</a> amidst continued adoption, according to new research. </p><p>Findings from Heimdal’s <em>State of AI Risk Management in 2026</em> report show <a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> are now commonplace across most IT estates, with teams often running several at one. </p><p><a href="https://www.itpro.com/technology/artificial-intelligence-ai/369965/what-is-chatgpt-and-what-does-it-mean-for-businesses">ChatGPT</a>, for example, runs in nearly three-quarters (71%) of UK IT environments, while <a href="https://www.itpro.com/technology/artificial-intelligence/microsoft-copilot-review-ai-baked-into-your-apps">Microsoft Copilot</a> is present in 68%. These same figures are reflected across the Atlantic, with US-based IT teams often using a combination of multiple different AI solutions. </p><p>AI tools are playing a key role in reducing workloads, the study noted, which IT teams highlighted as the most positive benefits of the technology. </p><p>Nearly three-quarters of IT and security teams said they lose around a quarter of their week to “repetitive, low-value work” - and AI is helping reduce that manual toil. </p><p>Indeed, teams facing the highest levels of operational load are often ranked among the most optimistic when it comes to AI. More than half (59%) of US teams said they expect AI to alleviate pressure, while 55% in the UK expect the same. </p><h2 id="ai-related-risks-are-haunting-it-teams">AI-related risks are haunting IT teams</h2><p>Despite this optimism, a key recurring concern among IT leaders is that controls and security capabilities haven’t kept pace with the rate of adoption. Heimdal noted that only four-in-ten teams rate their security stack as “ready for AI-related risk”. </p><p>Teams are increasingly concerned about data leakage, for example, with 56% of UK respondents highlighting this risk. Visibility is a major issue in this regard, Heimdal found. </p><p>UK teams with full <a href="https://www.itpro.com/security/it-leaders-are-facing-major-work-device-blind-spots-and-its-putting-security-at-risk">visibility into AI use</a> were most likely to flag data leakage as a leading concern, compared to just 27% of those with no visibility. In the US, 59% of teams with full visibility also highlighted data leakage as a key concern. </p><p>While concerns over potential risks typically mount as AI tools are integrated, Heimdal noted that unauthorized AI use, or ‘shadow AI’, is also a recurring problem for enterprises. </p><p>Heimdal specifically highlighted the <a href="https://www.itpro.com/security/cyber-attacks/the-salesloft-hackers-claim-they-have-1-5-billion-compromised-salesforce-records">Salesloft Drift breach in August 2025</a> as a key example of how poor AI-related visibility can impact organisations. </p><p>The incident saw threat actors steal OAuth tokens for Drift’s AI chatbot integration with Salesforce, using these to extract data from several hundred Salesforce instances. </p><p>A host of organisations, including Cloudflare, Palo Alto Networks, and Zscaler were impacted in the attacks.</p><p>“Drift was the AI tool. Salesforce held the data,” the company noted in a <a href="https://heimdalsecurity.com/blog/state-ai-risk-management/" target="_blank"><u>blog post</u></a>.</p><p>“Most of the affected teams had never personally provisioned Drift,” it added. “A third-party AI chatbot, plugged in through an OAuth grant few had recently reviewed, became the way in.”</p><h2 id="contrasting-priorities">Contrasting priorities</h2><p>Perception of AI-related risk among frontline practitioners and executives is a major problem, according to Heimdal. Indeed, “executive confidence” in AI security is a repeated point of friction when it comes to governance and risk management. </p><p>In the US, for example, 29% of executives said AI risk is under control, yet just 7% of practitioners agreed. In the UK, meanwhile, these figures stand at 18% against 11%. </p><p>Adam Pilton, cybersecurity advisor at Heimdal, said this shows many organizations still aren’t fully aligned on how they manage AI risk. </p><p>"Misplaced confidence is one of the most dangerous things in security. This data shows executives are far more confident that AI risk is under control than the evidence supports. Most of the conversation right now is about productivity, when the bigger question is how AI can be turned against the business,” he said. </p><p>“The report shows the gap between how secure leaders feel and how secure they actually are.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Anthropic suspends Fable and Mythos systems "for all users" ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Anthropic has suspended most of its advanced AI models just days after the company made them public, after sanctions were imposed by the Trump administration. </p><p>On Friday, Anthropic models Fable and Mythos were hit with export controls by the Department of Commerce, barring foreign nationals from using either model. In response, Anthropic has suspended both systems for all users. </p><p>In a statement posted on its website, Anthropic said that it was complying with the government's legal directive and removing access to <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-just-launched-claude-fable-5-its-first-mythos-class-ai-model-but-it-has-new-safeguards-to-prevent-misuse-and-will-fall-back-to-opus-4-8-for-high-risk-queries">Fable 5 and Mythos 5</a> for all users. But it disagreed with the government's reasoning as it only offered verbal evidence of a "potential narrow, non-universal jailbreak", according to Anthropic. </p><p>For AI systems, a <a href="https://www.itpro.com/technology/artificial-intelligence/some-of-the-most-popular-open-weight-ai-models-show-profound-susceptibility-to-jailbreak-techniques">jailbreak</a> is a technique used to manipulate an AI model into bypassing its built-in safeguards, guardrails, and content filters. An <a href="https://www.itpro.com/technology/artificial-intelligence/so-you-introduced-an-ai-chatbot-for-customers-heres-why-they-probably-hate-it">AI chatbot</a>, for example, which is designed to assist, has vulnerabilities that can be exploited by malicious actors and forced to do things it was specifically programmed not to. </p><p>Anthropic believes the jailbreak method claimed to have been found in Fable and or Mythos is widely available in other models, such as <a href="https://www.itpro.com/security/ai-is-getting-better-at-security-and-its-doing-it-faster-than-expected">OpenAI's GPT-5.5</a>, and is used daily by the security teams that keep it safe. </p><p>"We disagree that the finding of a narrow potential jailbreak should be a cause for recalling a commercial model deployed to hundreds of millions of people," the statement reads. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."</p><p>"Perfect" jailbreak resistance, according to Anthropic, is not currently possible for any model provider. And that every safeguard used in the industry is vulnerable to non-universal jailbreaks. The company added that it was likely that universal jailbreaks would eventually be found in the future, a point it stressed when it first released Fable 5.</p><h2 id="pentagon-aws-and-anthropic">Pentagon, AWS, and Anthropic</h2><p>The sanctions on Anthropic come not long after the Pentagon named the company as a supply-chain risk to national security. A number of senior government figures publicly clashed with Anthropic's leadership over issues including AI regulation and the use of Anthropic's technology in domestic surveillance and autonomous weapons. </p><p>The specific directive, however, followed a report by Amazon researchers who identified the potential jailbreak in Fable. <a href="https://www.ft.com/content/f6940d59-28f4-4ae4-a569-c6fc421e52b9?syn-25a6b1a6=1">Reportedly</a>, Amazon first shared its findings with Anthropic. CEO Andy Jassy is also said to have discussed the issue with US officials, raising broader concerns about frontier-model capabilities rather than issues focused specifically on Anthropic. </p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/anthropic-suspends-fabel-and-mythos-systems-for-all-users-after-us-government-claims-jailbreak-risk</link>
                                                                            <description>
                            <![CDATA[ Despite complying with the government, Anthropic suggests it's only a "potential narrow, non-universal jailbreak" ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">CDYt7RNcF6PmHRZ9zfKw79</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/B5sequrZU2Sim7e6k6gTnH-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Mon, 15 Jun 2026 09:36:42 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:06 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ itpro@futurenet.com (Bobby Hellard) ]]></author>                    <dc:creator><![CDATA[ Bobby Hellard ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/bsR2tHSyVKUoyXZF5pNsDA-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Bobby Hellard&amp;nbsp;is&amp;nbsp;ITPro&#039;s Reviews Editor and has worked on&amp;nbsp;CloudPro and ChannelPro since 2018. In his time at ITPro, Bobby has covered stories for all the major technology companies, such as Apple, Microsoft, Amazon and Facebook, and regularly attends industry-leading events such as AWS Re:Invent and Google Cloud Next.&lt;/p&gt;
&lt;p&gt;Bobby mainly covers hardware reviews, but you will also recognize him as the face of many of our video reviews of laptops and smartphones.&lt;/p&gt;
&lt;p&gt;He has been a journalist for ten years, originally covering sports, before moving into business technology with ITPro. He has bylines in The Independent, Vice and The Business Briefing. Contact him at &lt;a href=&quot;mailto:bobby.hellard@futurenet.com&quot;&gt;bobby.hellard@futurenet.com&lt;/a&gt; or find him on Twitter: &lt;a href=&quot;https://twitter.com/bobbyhellard&quot;&gt;@bobbyhellard&lt;/a&gt;&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/B5sequrZU2Sim7e6k6gTnH-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[The Anthropic logo, in white on a dark background lit by a web-like light.]]></media:description>                                                            <media:text><![CDATA[The Anthropic logo, in white on a dark background lit by a web-like light.]]></media:text>
                                <media:title type="plain"><![CDATA[The Anthropic logo, in white on a dark background lit by a web-like light.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/B5sequrZU2Sim7e6k6gTnH-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Anthropic has suspended most of its advanced AI models just days after the company made them public, after sanctions were imposed by the Trump administration. </p><p>On Friday, Anthropic models Fable and Mythos were hit with export controls by the Department of Commerce, barring foreign nationals from using either model. In response, Anthropic has suspended both systems for all users. </p><p>In a statement posted on its website, Anthropic said that it was complying with the government's legal directive and removing access to <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-just-launched-claude-fable-5-its-first-mythos-class-ai-model-but-it-has-new-safeguards-to-prevent-misuse-and-will-fall-back-to-opus-4-8-for-high-risk-queries">Fable 5 and Mythos 5</a> for all users. But it disagreed with the government's reasoning as it only offered verbal evidence of a "potential narrow, non-universal jailbreak", according to Anthropic. </p><p>For AI systems, a <a href="https://www.itpro.com/technology/artificial-intelligence/some-of-the-most-popular-open-weight-ai-models-show-profound-susceptibility-to-jailbreak-techniques">jailbreak</a> is a technique used to manipulate an AI model into bypassing its built-in safeguards, guardrails, and content filters. An <a href="https://www.itpro.com/technology/artificial-intelligence/so-you-introduced-an-ai-chatbot-for-customers-heres-why-they-probably-hate-it">AI chatbot</a>, for example, which is designed to assist, has vulnerabilities that can be exploited by malicious actors and forced to do things it was specifically programmed not to. </p><p>Anthropic believes the jailbreak method claimed to have been found in Fable and or Mythos is widely available in other models, such as <a href="https://www.itpro.com/security/ai-is-getting-better-at-security-and-its-doing-it-faster-than-expected">OpenAI's GPT-5.5</a>, and is used daily by the security teams that keep it safe. </p><p>"We disagree that the finding of a narrow potential jailbreak should be a cause for recalling a commercial model deployed to hundreds of millions of people," the statement reads. "If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers."</p><p>"Perfect" jailbreak resistance, according to Anthropic, is not currently possible for any model provider. And that every safeguard used in the industry is vulnerable to non-universal jailbreaks. The company added that it was likely that universal jailbreaks would eventually be found in the future, a point it stressed when it first released Fable 5.</p><h2 id="pentagon-aws-and-anthropic">Pentagon, AWS, and Anthropic</h2><p>The sanctions on Anthropic come not long after the Pentagon named the company as a supply-chain risk to national security. A number of senior government figures publicly clashed with Anthropic's leadership over issues including AI regulation and the use of Anthropic's technology in domestic surveillance and autonomous weapons. </p><p>The specific directive, however, followed a report by Amazon researchers who identified the potential jailbreak in Fable. <a href="https://www.ft.com/content/f6940d59-28f4-4ae4-a569-c6fc421e52b9?syn-25a6b1a6=1">Reportedly</a>, Amazon first shared its findings with Anthropic. CEO Andy Jassy is also said to have discussed the issue with US officials, raising broader concerns about frontier-model capabilities rather than issues focused specifically on Anthropic. </p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Closing the AI control gap: Why channel partners are now on the front line   ]]></title>
                                                                                                <dc:content><![CDATA[ <p>In recent years, for many MSPs and MSSPs, AI has moved quickly from a future conversation to a current customer problem. Indeed, our own <a href="https://www.culture.ai/resources/whitepapers/the-state-of-enterprise-ai-usage-control"><u>research</u></a> found that 72% of organizations believe they have full visibility into AI usage, yet 65% are still uncovering shadow or unauthorised AI activity. </p><p>Most businesses, particularly those that operate in highly regulated sectors, acknowledge the risks of AI, but are unaware of how to control it. As a result, many organizations are now turning to partners to help them bridge the gap between perceived confidence and reality. </p><p>While this challenge presents a significant commercial opportunity for channel partners, it also highlights the importance of understanding the risks of shadow AI and uncontrolled usage and how they can effectively advise customers to harness this emerging opportunity. </p><h2 id="understanding-the-ai-control-gap-for-businesses">Understanding the AI control gap for businesses </h2><p>The core challenge organizations currently face is a misunderstanding of how AI is actually being used within the business. The popularity of generative AI tools such as large language models (LLMs) has skyrocketed, as organizations look to boost productivity, with a reported <a href="https://evolvagency.io/learn/generative-search/llm-statistics-2026"><u>92% of Fortune 100 companies</u></a> now using ChatGPT in some capacity. Yet the risk lies in how these tools are being used by employees, which is often without formal oversight and with limited awareness of the potential risks involved.</p><p>While the instinct for business leaders to ban AI use entirely is understandable, it's completely counterproductive. Employees who have already benefited from these tools will not simply stop using them; they will inevitably find workarounds, often on personal devices with no organizational visibility whatsoever. A blanket ban does not reduce risk; it simply relocates it beyond reach. Here, the role of channel partners, particularly MSPs and MSSPs, becomes not just relevant but essential. </p><p>Organizations need guidance from advisors they already trust, and the skills required to close the AI control gap sit squarely within the channel's wheelhouse. Therefore, partners need to leverage technical and strategic skills in this area to help customers tackle this issue. </p><h2 id="why-ai-governance-is-the-channel-s-next-big-opportunity">Why AI governance Is the channel's next big opportunity </h2><p>For many channel partners, the starting point is shifting how customers think about AI risk and visibility. Having a policy in place is not the same as having control, yet, according to our research, more than 60% of organizations already believe they have adequate frameworks and oversight. In most cases, that confidence is misplaced, because policy without genuine visibility is not governance but assumption dressed up as oversight. </p><p>The reality is that AI usage is fragmented, spreading across teams, devices, and applications in ways that make a complete picture almost impossible to build through traditional means. </p><p>Sanctioned tools sit alongside unsanctioned ones, and employee behaviour routinely outpaces the policies designed to govern it. Helping customers see that gap clearly and understand what it actually exposes them to is where channel partners can add immediate value.</p><p>The customer demand to support that guidance is already well established, with many organizations actively seeking answers to questions around which tools are safe to use, how sensitive data should be handled, and how employees can work with AI effectively. As a result, the demand for advisory engagements around AI governance, risk assessments, and policy development growing in value, as well as continuous monitoring and behavioural analysis to detect risky usage in real time, is growing in value. </p><p>This presents a significant commercial opportunity for the channel by unlocking the potential for new services and revenue streams. Another advantage is that AI usage control is not a project with a defined end date, but rather an ongoing service discipline that plays directly to the strengths of MSPs and MSSPs built for managed, long-term customer relationships.</p><h2 id="what-effective-ai-usage-control-looks-like">What effective AI usage control looks like </h2><p>The key to providing effective support to organizations is looking beyond technical solutions alone and striking the right balance between security and visibility to effectively govern AI usage. This is particularly important given that overly restrictive policies can actively undermine those efforts, consequently pushing employees toward shadow tools and increasing the high risks organizations are trying to reduce. </p><p>The main focus should be directed towards educating organizations to help them understand how AI is being used across the business, identifying high-risk behaviours, and guiding employees in real time. Channel partners play a critical role in assisting organizations with embedding safe AI usage into everyday workflows and becoming a natural part of how they operate. </p><p>In the long run, the AI control gap is not a failure of technology but a reflection of how quickly adoption has outpaced governance. While businesses are right to prioritise the benefits of AI, it is equally important that they do so within a framework of guardrails that ensure it is used safely.  </p><p>As that gap continues to widen, channel partners who combine technical depth with strategic insight will be best placed to lead, helping customers move from guesswork to control, and turning AI from an unmanaged risk into a managed advantage. </p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/closing-the-ai-control-gap-why-channel-partners-are-now-on-the-front-line</link>
                                                                            <description>
                            <![CDATA[ AI adoption creates unmanaged risks, driving demand for expert partner guidance ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">eaWKgbptUfKq9rtuwbGKVD</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/bZPUo6nPPSTtRxt3UUME8d-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 28 May 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:48 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Ryan Davis ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/ArRMYPtzxwCzUkdJaU7CV4-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/bZPUo6nPPSTtRxt3UUME8d-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Green AI symbol illuminated against a dark backdrop with message dashboard symbols surrounding symbol.]]></media:description>                                                            <media:text><![CDATA[Green AI symbol illuminated against a dark backdrop with message dashboard symbols surrounding symbol.]]></media:text>
                                <media:title type="plain"><![CDATA[Green AI symbol illuminated against a dark backdrop with message dashboard symbols surrounding symbol.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/bZPUo6nPPSTtRxt3UUME8d-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>In recent years, for many MSPs and MSSPs, AI has moved quickly from a future conversation to a current customer problem. Indeed, our own <a href="https://www.culture.ai/resources/whitepapers/the-state-of-enterprise-ai-usage-control"><u>research</u></a> found that 72% of organizations believe they have full visibility into AI usage, yet 65% are still uncovering shadow or unauthorised AI activity. </p><p>Most businesses, particularly those that operate in highly regulated sectors, acknowledge the risks of AI, but are unaware of how to control it. As a result, many organizations are now turning to partners to help them bridge the gap between perceived confidence and reality. </p><p>While this challenge presents a significant commercial opportunity for channel partners, it also highlights the importance of understanding the risks of shadow AI and uncontrolled usage and how they can effectively advise customers to harness this emerging opportunity. </p><h2 id="understanding-the-ai-control-gap-for-businesses">Understanding the AI control gap for businesses </h2><p>The core challenge organizations currently face is a misunderstanding of how AI is actually being used within the business. The popularity of generative AI tools such as large language models (LLMs) has skyrocketed, as organizations look to boost productivity, with a reported <a href="https://evolvagency.io/learn/generative-search/llm-statistics-2026"><u>92% of Fortune 100 companies</u></a> now using ChatGPT in some capacity. Yet the risk lies in how these tools are being used by employees, which is often without formal oversight and with limited awareness of the potential risks involved.</p><p>While the instinct for business leaders to ban AI use entirely is understandable, it's completely counterproductive. Employees who have already benefited from these tools will not simply stop using them; they will inevitably find workarounds, often on personal devices with no organizational visibility whatsoever. A blanket ban does not reduce risk; it simply relocates it beyond reach. Here, the role of channel partners, particularly MSPs and MSSPs, becomes not just relevant but essential. </p><p>Organizations need guidance from advisors they already trust, and the skills required to close the AI control gap sit squarely within the channel's wheelhouse. Therefore, partners need to leverage technical and strategic skills in this area to help customers tackle this issue. </p><h2 id="why-ai-governance-is-the-channel-s-next-big-opportunity">Why AI governance Is the channel's next big opportunity </h2><p>For many channel partners, the starting point is shifting how customers think about AI risk and visibility. Having a policy in place is not the same as having control, yet, according to our research, more than 60% of organizations already believe they have adequate frameworks and oversight. In most cases, that confidence is misplaced, because policy without genuine visibility is not governance but assumption dressed up as oversight. </p><p>The reality is that AI usage is fragmented, spreading across teams, devices, and applications in ways that make a complete picture almost impossible to build through traditional means. </p><p>Sanctioned tools sit alongside unsanctioned ones, and employee behaviour routinely outpaces the policies designed to govern it. Helping customers see that gap clearly and understand what it actually exposes them to is where channel partners can add immediate value.</p><p>The customer demand to support that guidance is already well established, with many organizations actively seeking answers to questions around which tools are safe to use, how sensitive data should be handled, and how employees can work with AI effectively. As a result, the demand for advisory engagements around AI governance, risk assessments, and policy development growing in value, as well as continuous monitoring and behavioural analysis to detect risky usage in real time, is growing in value. </p><p>This presents a significant commercial opportunity for the channel by unlocking the potential for new services and revenue streams. Another advantage is that AI usage control is not a project with a defined end date, but rather an ongoing service discipline that plays directly to the strengths of MSPs and MSSPs built for managed, long-term customer relationships.</p><h2 id="what-effective-ai-usage-control-looks-like">What effective AI usage control looks like </h2><p>The key to providing effective support to organizations is looking beyond technical solutions alone and striking the right balance between security and visibility to effectively govern AI usage. This is particularly important given that overly restrictive policies can actively undermine those efforts, consequently pushing employees toward shadow tools and increasing the high risks organizations are trying to reduce. </p><p>The main focus should be directed towards educating organizations to help them understand how AI is being used across the business, identifying high-risk behaviours, and guiding employees in real time. Channel partners play a critical role in assisting organizations with embedding safe AI usage into everyday workflows and becoming a natural part of how they operate. </p><p>In the long run, the AI control gap is not a failure of technology but a reflection of how quickly adoption has outpaced governance. While businesses are right to prioritise the benefits of AI, it is equally important that they do so within a framework of guardrails that ensure it is used safely.  </p><p>As that gap continues to widen, channel partners who combine technical depth with strategic insight will be best placed to lead, helping customers move from guesswork to control, and turning AI from an unmanaged risk into a managed advantage. </p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ UK and Australia agree to work more closely on AI security ]]></title>
                                                                                                <dc:content><![CDATA[ <p>The UK and Australia have announced plans to extend their cooperation on AI security and safety.</p><p>The two plan to share insights on AI capabilities, carry out research into emerging risks, and work together to develop international best practice for testing and evaluating AI systems. They plan to carry out joint research, including, they said, new approaches to the measurement, testing and management of risks.</p><p>"Australia and the UK have always worked closely to keep our people safe – and that partnership matters more than ever in the <a href="https://www.itpro.com/security/in-the-age-of-ai-threats-the-future-of-security-is-unified">age of AI</a>. This technology is moving fast, and so are the risks that come with it – particularly in areas like <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity</a>. No country can tackle that alone," said UK AI minister Kanishka Narayan.</p><p>"By working hand in hand with trusted partners such as Australia, we can stay ahead of the risks, strengthen our defences and make sure <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI</a> is used to improve lives for Brits and Aussies alike."</p><p>The new Memorandum of Understanding (MoU) between the two countries is based on a new partnership between the UK AI Security Institute and a newly-created Australian AI Safety Institute. This, said the government, will monitor, test, and share information on emerging AI capabilities, risks, and harms, making sure that Australia has the capability to keep up with developments in AI technologies. And the new deal will see the two institutes working closely, providing for staff exchanges to strengthen day-to-day collaboration. </p><p>"This MoU brings Australia's AI Safety Institute and the UK AI Security Institute together to share expertise, identify risks early and support safeguards that help prevent harm," said Australian minister for industry and innovation and minister for science Tim Ayres.</p><p>"Collaboration with our closest partners is how we make sure AI is safe and secure while we continue to explore the opportunities it presents."</p><p>AISI already has similar agreements with AI research bodies across the world's major economies, developed through the International Network for Advanced AI Measurement, Evaluation, and Science, and through its bilateral partnerships.</p><p>And this deal builds on a 2024 MoU between the UK and Australia on online safety and security covering illegal content, child safety, age assurance, technology-facilitated gender-based violence, and the harms of generative AI.</p><p>It committed the two countries to in-person dialogue, coordinated bilateral and multilateral engagement, regulatory engagement, shared research projects, and working with industry to address safety challenges.</p><p>"This agreement with the UK reflects our shared commitment to ensuring artificial intelligence is developed and deployed safely, securely and responsibly," said Australia's assistant minister for science, technology and the digital economy, Dr Andrew Charlton.</p><p>"By strengthening cooperation with democracies, we are combining world-class expertise to understand emerging risks and keep pace with rapid technological change."</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/uk-and-australia-agree-to-work-more-closely-on-ai-security</link>
                                                                            <description>
                            <![CDATA[ A new deal sees Australia set up a new AI safety institute, which will share research with the UK AI Security Institute ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">jH4XJQtpVojVB5K67Ezma7</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/CoDazdKR4EHWNz2Kf7x9rf-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 26 May 2026 10:01:24 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:02 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/CoDazdKR4EHWNz2Kf7x9rf-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[A digital Australian flag ]]></media:description>                                                            <media:text><![CDATA[A digital Australian flag ]]></media:text>
                                <media:title type="plain"><![CDATA[A digital Australian flag ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/CoDazdKR4EHWNz2Kf7x9rf-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>The UK and Australia have announced plans to extend their cooperation on AI security and safety.</p><p>The two plan to share insights on AI capabilities, carry out research into emerging risks, and work together to develop international best practice for testing and evaluating AI systems. They plan to carry out joint research, including, they said, new approaches to the measurement, testing and management of risks.</p><p>"Australia and the UK have always worked closely to keep our people safe – and that partnership matters more than ever in the <a href="https://www.itpro.com/security/in-the-age-of-ai-threats-the-future-of-security-is-unified">age of AI</a>. This technology is moving fast, and so are the risks that come with it – particularly in areas like <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity</a>. No country can tackle that alone," said UK AI minister Kanishka Narayan.</p><p>"By working hand in hand with trusted partners such as Australia, we can stay ahead of the risks, strengthen our defences and make sure <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI</a> is used to improve lives for Brits and Aussies alike."</p><p>The new Memorandum of Understanding (MoU) between the two countries is based on a new partnership between the UK AI Security Institute and a newly-created Australian AI Safety Institute. This, said the government, will monitor, test, and share information on emerging AI capabilities, risks, and harms, making sure that Australia has the capability to keep up with developments in AI technologies. And the new deal will see the two institutes working closely, providing for staff exchanges to strengthen day-to-day collaboration. </p><p>"This MoU brings Australia's AI Safety Institute and the UK AI Security Institute together to share expertise, identify risks early and support safeguards that help prevent harm," said Australian minister for industry and innovation and minister for science Tim Ayres.</p><p>"Collaboration with our closest partners is how we make sure AI is safe and secure while we continue to explore the opportunities it presents."</p><p>AISI already has similar agreements with AI research bodies across the world's major economies, developed through the International Network for Advanced AI Measurement, Evaluation, and Science, and through its bilateral partnerships.</p><p>And this deal builds on a 2024 MoU between the UK and Australia on online safety and security covering illegal content, child safety, age assurance, technology-facilitated gender-based violence, and the harms of generative AI.</p><p>It committed the two countries to in-person dialogue, coordinated bilateral and multilateral engagement, regulatory engagement, shared research projects, and working with industry to address safety challenges.</p><p>"This agreement with the UK reflects our shared commitment to ensuring artificial intelligence is developed and deployed safely, securely and responsibly," said Australia's assistant minister for science, technology and the digital economy, Dr Andrew Charlton.</p><p>"By strengthening cooperation with democracies, we are combining world-class expertise to understand emerging risks and keep pace with rapid technological change."</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Trump wants AI models 90 days before they go public ]]></title>
                                                                                                <dc:content><![CDATA[ <p>AI developers could be forced to hand over their new frontier models 90 days before a public launch, according to leaked accounts of a draft executive order from US President Donald Trump.</p><p>Multiple media reports have cited sources saying that the Trump administration plans to issue an executive order as early as this week, laying out government agencies that will get advanced access to <a href="https://www.itpro.com/software/development/frontier-models-are-still-unable-to-solve-the-majority-of-tasks-ai-might-not-replace-software-engineers-just-yet-openai-researchers-found-leading-models-and-coding-tools-still-lag-behind-humans-on-basic-tasks">frontier models</a> in order to scrutinise their capabilities. </p><p>Reports suggest the administration could be demanding access to new frontier models 90 days before they're made public, according to <a href="https://www.itpro.com/software/development/frontier-models-are-still-unable-to-solve-the-majority-of-tasks-ai-might-not-replace-software-engineers-just-yet-openai-researchers-found-leading-models-and-coding-tools-still-lag-behind-humans-on-basic-tasks"><em>Axios</em></a>, and give the Pentagon time to look over security-focused models – and secure its own systems, as well as alert other impacted organisations of potentially looming flaws. <a href="https://www.reuters.com/legal/litigation/trump-sign-order-ai-oversight-security-fears-mount-among-supporters-2026-05-20/"><em>Reuters</em></a> said a source suggested critical infrastructure organisations such as banks could also get access to frontier models before the general public.</p><p>Beyond the advance access, the order will ask AI developers to keep the government updated on incoming models and offer access to their infrastructure. </p><p>According to a report from <a href="https://www.politico.com/news/2026/05/20/trump-ai-order-details-00930681"><em>Politico</em></a>, the White House began detailing the directive to AI companies on Tuesday, and it's expected to be made official as early as today. The Pentagon is expected to be given 30 days to set up its security "clearinghouse", while a group of federal agencies will have 60 days to decide what defines a model as "frontier". </p><p>A White House spokesperson told <a href="https://www.reuters.com/legal/litigation/trump-sign-order-ai-oversight-security-fears-mount-among-supporters-2026-05-20/"><em>Reuters</em></a> the reports were nothing but "speculation." Last year, Trump issued an executive order banning states from local regulation of AI in order to create a federal rule. </p><h2 id="drawn-out-process">Drawn out process </h2><p>The details of this latest executive order have been much debated and have changed multiple times throughout the process, with one <a href="https://www.nextgov.com/artificial-intelligence/2026/05/anticipated-executive-order-could-give-nsa-role-voluntary-ai-model-testing/413663/">report</a> noting that the order could change or be further delayed if discussions with AI labs don't go as planned. </p><p><a href="https://edition.cnn.com/2026/05/20/tech/ai-executive-order-trump-white-house"><em>CNN</em></a> noted that some of the AI companies involved in the discussion have pushed back against the 90-day access in favour of just 14 days. On the other hand, a group of Republican politicians wrote to Trump <a href="https://www.reuters.com/legal/litigation/trump-sign-order-ai-oversight-security-fears-mount-among-supporters-2026-05-20/">last week</a> demanding government oversight and approval of "potentially dangerous" systems. </p><p>Key industry players, including Google and Microsoft, two weeks ago agreed to give the government early access via the Centre for AI Standards and Innovation at the Department of Commerce, but – as <a href="https://www.reuters.com/legal/litigation/microsoft-google-xai-security-test-details-deleted-us-government-website-2026-05-11/"><em>Reuters</em></a> notes – that announcement disappeared from the department's website within days. </p><p>The long-expected executive order follows concerns raised by <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-lets-glasswing-partners-publicly-share-mythos-flaws">Anthropic's Mythos security model</a>, as well as that company's break with the Pentagon over military uses of its AI. </p><h2 id="ai-layoffs">AI layoffs</h2><p>It also comes amid layoffs that companies have pinned on AI, including thousands impacted at <a href="https://www.itpro.com/business/leadership/internal-memo-suggests-meta-will-lay-off-10-percent-of-its-employees-with-a-further-10-percent-set-to-be-transferred-to-better-focus-on-ai">Meta this week</a>, highlighting the difficulty politicians have in walking a line showing support for AI as a technology that could one day boost the economy while causing disruption now, be it security concerns, data centre complaints or unemployment. </p><p>Asked this week what message he had for Americans concerned about AI's impact on their jobs, President Trump <a href="https://uk.news.yahoo.com/trump-says-ai-amazing-while-203145592.html">said</a>: "I'll tell you, AI has been amazing because we have more jobs, more people working right now in the United States by far than we ever had before." </p><p>Earlier this year, Trump agreed <a href="https://www.itpro.com/technology/artificial-intelligence/trump-makes-ai-firms-pay-for-power">a deal with Big Tech companies</a> designed to ensure the rollout of data centres and other AI infrastructure doesn't drive up electricity prices in the US.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/trump-wants-ai-models-90-days-before-they-go-public</link>
                                                                            <description>
                            <![CDATA[ Latest attempt to regulate AI could come this week as the Trump administration demands early access to frontier models ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">z8aWxGWXNnzkfPYZdsSYFL</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/c46xyh4g5ChSfCCg3ffePi-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 21 May 2026 10:47:05 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:08 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/c46xyh4g5ChSfCCg3ffePi-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[ US President Donald Trump speaks during a proclamation signing in the Oval Office of the White House on Monday August 25, 2025.]]></media:description>                                                            <media:text><![CDATA[ US President Donald Trump speaks during a proclamation signing in the Oval Office of the White House on Monday August 25, 2025.]]></media:text>
                                <media:title type="plain"><![CDATA[ US President Donald Trump speaks during a proclamation signing in the Oval Office of the White House on Monday August 25, 2025.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/c46xyh4g5ChSfCCg3ffePi-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>AI developers could be forced to hand over their new frontier models 90 days before a public launch, according to leaked accounts of a draft executive order from US President Donald Trump.</p><p>Multiple media reports have cited sources saying that the Trump administration plans to issue an executive order as early as this week, laying out government agencies that will get advanced access to <a href="https://www.itpro.com/software/development/frontier-models-are-still-unable-to-solve-the-majority-of-tasks-ai-might-not-replace-software-engineers-just-yet-openai-researchers-found-leading-models-and-coding-tools-still-lag-behind-humans-on-basic-tasks">frontier models</a> in order to scrutinise their capabilities. </p><p>Reports suggest the administration could be demanding access to new frontier models 90 days before they're made public, according to <a href="https://www.itpro.com/software/development/frontier-models-are-still-unable-to-solve-the-majority-of-tasks-ai-might-not-replace-software-engineers-just-yet-openai-researchers-found-leading-models-and-coding-tools-still-lag-behind-humans-on-basic-tasks"><em>Axios</em></a>, and give the Pentagon time to look over security-focused models – and secure its own systems, as well as alert other impacted organisations of potentially looming flaws. <a href="https://www.reuters.com/legal/litigation/trump-sign-order-ai-oversight-security-fears-mount-among-supporters-2026-05-20/"><em>Reuters</em></a> said a source suggested critical infrastructure organisations such as banks could also get access to frontier models before the general public.</p><p>Beyond the advance access, the order will ask AI developers to keep the government updated on incoming models and offer access to their infrastructure. </p><p>According to a report from <a href="https://www.politico.com/news/2026/05/20/trump-ai-order-details-00930681"><em>Politico</em></a>, the White House began detailing the directive to AI companies on Tuesday, and it's expected to be made official as early as today. The Pentagon is expected to be given 30 days to set up its security "clearinghouse", while a group of federal agencies will have 60 days to decide what defines a model as "frontier". </p><p>A White House spokesperson told <a href="https://www.reuters.com/legal/litigation/trump-sign-order-ai-oversight-security-fears-mount-among-supporters-2026-05-20/"><em>Reuters</em></a> the reports were nothing but "speculation." Last year, Trump issued an executive order banning states from local regulation of AI in order to create a federal rule. </p><h2 id="drawn-out-process">Drawn out process </h2><p>The details of this latest executive order have been much debated and have changed multiple times throughout the process, with one <a href="https://www.nextgov.com/artificial-intelligence/2026/05/anticipated-executive-order-could-give-nsa-role-voluntary-ai-model-testing/413663/">report</a> noting that the order could change or be further delayed if discussions with AI labs don't go as planned. </p><p><a href="https://edition.cnn.com/2026/05/20/tech/ai-executive-order-trump-white-house"><em>CNN</em></a> noted that some of the AI companies involved in the discussion have pushed back against the 90-day access in favour of just 14 days. On the other hand, a group of Republican politicians wrote to Trump <a href="https://www.reuters.com/legal/litigation/trump-sign-order-ai-oversight-security-fears-mount-among-supporters-2026-05-20/">last week</a> demanding government oversight and approval of "potentially dangerous" systems. </p><p>Key industry players, including Google and Microsoft, two weeks ago agreed to give the government early access via the Centre for AI Standards and Innovation at the Department of Commerce, but – as <a href="https://www.reuters.com/legal/litigation/microsoft-google-xai-security-test-details-deleted-us-government-website-2026-05-11/"><em>Reuters</em></a> notes – that announcement disappeared from the department's website within days. </p><p>The long-expected executive order follows concerns raised by <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-lets-glasswing-partners-publicly-share-mythos-flaws">Anthropic's Mythos security model</a>, as well as that company's break with the Pentagon over military uses of its AI. </p><h2 id="ai-layoffs">AI layoffs</h2><p>It also comes amid layoffs that companies have pinned on AI, including thousands impacted at <a href="https://www.itpro.com/business/leadership/internal-memo-suggests-meta-will-lay-off-10-percent-of-its-employees-with-a-further-10-percent-set-to-be-transferred-to-better-focus-on-ai">Meta this week</a>, highlighting the difficulty politicians have in walking a line showing support for AI as a technology that could one day boost the economy while causing disruption now, be it security concerns, data centre complaints or unemployment. </p><p>Asked this week what message he had for Americans concerned about AI's impact on their jobs, President Trump <a href="https://uk.news.yahoo.com/trump-says-ai-amazing-while-203145592.html">said</a>: "I'll tell you, AI has been amazing because we have more jobs, more people working right now in the United States by far than we ever had before." </p><p>Earlier this year, Trump agreed <a href="https://www.itpro.com/technology/artificial-intelligence/trump-makes-ai-firms-pay-for-power">a deal with Big Tech companies</a> designed to ensure the rollout of data centres and other AI infrastructure doesn't drive up electricity prices in the US.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ European Commission opens public consultation on draft for high-risk AI guidelines ]]></title>
                                                                                                <dc:content><![CDATA[ <p>The European Commission has published its long-awaited draft guidance on the classification of <a href="https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry">high-risk AI systems</a> and has opened a public consultation.</p><p>The <a href="https://digital-strategy.ec.europa.eu/en/library/draft-commission-guidelines-classification-high-risk-ai-systems">guidelines</a> aim to help assess whether an AI system should be classified as high-risk; either intended to be used as a safety component of a product, or carrying the potential to endanger health, safety or fundamental rights. If they do, they need different treatment under Article 6 of the <a href="https://www.itpro.com/business/policy-and-legislation/how-the-eu-ai-act-compares-to-other-international-regulatory-approaches">AI Act</a>.</p><p>"The guidelines are intended to support providers, deployers, and other relevant actors in determining whether an AI system falls within the high-risk category," the Commission explained. </p><p>"They offer clarifications on the relevant provisions of the <a href="https://www.itpro.com/technology/artificial-intelligence/eu-ai-act-everything-you-need-to-know-about-the-legislation-including-rules-requirements-and-who-will-be-forced-to-comply">AI Act</a> and include practical examples to illustrate how the classification should be assessed in different areas and use cases."</p><p>The guide comes in two parts, with the first covering AI systems that are themselves products, or safety components of products, under sectoral harmonization legislation, including the Machinery Regulation, the Toys Safety Regulation, the Radio Equipment Directive, medical devices, and the automotive and aviation regimes.</p><p>The second, meanwhile, covers AI systems that can significantly affect people's health, safety, or fundamental rights in specific use cases listed in the AI Act.</p><p>Article 6(3), covering the significant risk of harm to health, safety, or fundamental rights, is where most providers will go wrong, said Patrick Sullivan, VP of Strategy and Innovation at compliance provider A-LIGN.</p><p>There's a carve-out with four conditions – that the AI performs a narrow procedural task, improves the result of a previously completed human activity, detects decision-making patterns or deviations without replacing or influencing the human assessment, and performs a preparatory task to an Annex III assessment.</p><p>But, said Sullivan, hard exception applies, with an Annex III system is always high-risk where it performs profiling of natural persons. "This matters enormously for HR systems, credit decisioning, and any system that builds individual-level predictive profiles," he warned.</p><p>The document has been significantly delayed, having been expected to be published back in February; and following the political agreement on the AI Omnibus, there's a new enforcement timeline. </p><p>Rules for systems used in certain high-risk areas – including biometrics, critical infrastructure, education, employment, migration, asylum and border control – will apply from 2 December 2027, while for systems integrated into products such as robotics and industrial machinery, they'll come into effect on 2 August 2028.</p><p>The Commission is now holding a consultation on the guidelines, aimed at anyone with an interest in the development, deployment, supervision or use of AI systems is invited to contribute. </p><p>"This includes AI providers and developers, organisations using AI systems, public authorities, researchers, civil society organisations, supervisory bodies and members of the public," it explained.</p><p>The consultation is open until 23 June.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/policy-and-legislation/european-commission-opens-public-consultation-on-long-awaited-draft-for-high-risk-ai-guidelines</link>
                                                                            <description>
                            <![CDATA[ Guidance aims to help organizations and regulators decide whether their AI products and deployments need to conform to tougher rules ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">3wo6mFPTKWFyxRVUWbjZLF</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/qnGSwbfzp9zTsFt2t49oUT-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 20 May 2026 11:08:58 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:54 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Policy and Legislation]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/qnGSwbfzp9zTsFt2t49oUT-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[EU flags fly outside the union headquarters in Brussels, Belgium.]]></media:description>                                                            <media:text><![CDATA[EU flags fly outside the union headquarters in Brussels, Belgium.]]></media:text>
                                <media:title type="plain"><![CDATA[EU flags fly outside the union headquarters in Brussels, Belgium.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/qnGSwbfzp9zTsFt2t49oUT-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>The European Commission has published its long-awaited draft guidance on the classification of <a href="https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry">high-risk AI systems</a> and has opened a public consultation.</p><p>The <a href="https://digital-strategy.ec.europa.eu/en/library/draft-commission-guidelines-classification-high-risk-ai-systems">guidelines</a> aim to help assess whether an AI system should be classified as high-risk; either intended to be used as a safety component of a product, or carrying the potential to endanger health, safety or fundamental rights. If they do, they need different treatment under Article 6 of the <a href="https://www.itpro.com/business/policy-and-legislation/how-the-eu-ai-act-compares-to-other-international-regulatory-approaches">AI Act</a>.</p><p>"The guidelines are intended to support providers, deployers, and other relevant actors in determining whether an AI system falls within the high-risk category," the Commission explained. </p><p>"They offer clarifications on the relevant provisions of the <a href="https://www.itpro.com/technology/artificial-intelligence/eu-ai-act-everything-you-need-to-know-about-the-legislation-including-rules-requirements-and-who-will-be-forced-to-comply">AI Act</a> and include practical examples to illustrate how the classification should be assessed in different areas and use cases."</p><p>The guide comes in two parts, with the first covering AI systems that are themselves products, or safety components of products, under sectoral harmonization legislation, including the Machinery Regulation, the Toys Safety Regulation, the Radio Equipment Directive, medical devices, and the automotive and aviation regimes.</p><p>The second, meanwhile, covers AI systems that can significantly affect people's health, safety, or fundamental rights in specific use cases listed in the AI Act.</p><p>Article 6(3), covering the significant risk of harm to health, safety, or fundamental rights, is where most providers will go wrong, said Patrick Sullivan, VP of Strategy and Innovation at compliance provider A-LIGN.</p><p>There's a carve-out with four conditions – that the AI performs a narrow procedural task, improves the result of a previously completed human activity, detects decision-making patterns or deviations without replacing or influencing the human assessment, and performs a preparatory task to an Annex III assessment.</p><p>But, said Sullivan, hard exception applies, with an Annex III system is always high-risk where it performs profiling of natural persons. "This matters enormously for HR systems, credit decisioning, and any system that builds individual-level predictive profiles," he warned.</p><p>The document has been significantly delayed, having been expected to be published back in February; and following the political agreement on the AI Omnibus, there's a new enforcement timeline. </p><p>Rules for systems used in certain high-risk areas – including biometrics, critical infrastructure, education, employment, migration, asylum and border control – will apply from 2 December 2027, while for systems integrated into products such as robotics and industrial machinery, they'll come into effect on 2 August 2028.</p><p>The Commission is now holding a consultation on the guidelines, aimed at anyone with an interest in the development, deployment, supervision or use of AI systems is invited to contribute. </p><p>"This includes AI providers and developers, organisations using AI systems, public authorities, researchers, civil society organisations, supervisory bodies and members of the public," it explained.</p><p>The consultation is open until 23 June.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ UK public expecting AI-driven recession ]]></title>
                                                                                                <dc:content><![CDATA[ <p>The British public is deeply fearful of AI, with one-in-five even thinking it will lead to civil unrest.</p><p>New research by King's College London's Institute for Artificial Intelligence and Policy Institute has found that seven-in-ten people are worried about the economic impacts of AI, six-in-ten think it will eliminate more jobs than it creates, and half think its impact will be worse than a normal recession.</p><p>Almost half the public – 48% – said they'd rather avoid AI, while 41% are afraid of it. Only 24% think it's positive for humanity, compared with 39% who disagree.</p><p>There's a noticeable gender gap here, with women more likely to disagree than agree that AI is positive for humanity, at 44% versus 18%; positive for the UK (34% vs 23%; and that AI will improve their lives (41% vs 22%). </p><p>University students are more optimistic than the wider public, with 47% saying AI is positive for the UK and 43% that it is positive for humanity, compared with just 28% and 24% of the general public, respectively.</p><p>Meanwhile, seven-in-ten workers are worried about the economic impact of <a href="https://www.itpro.com/business/business-strategy/ai-job-losses-great-recession-us-forrester">AI job losses</a> – and the figure's almost as high for employers, at 64%. Most of the public (57%) thinks AI will lead to widespread unemployment.</p><p>People are particularly worried about the impacts on entry-level jobs and young people more generally, with nearly six-in-ten agreeing with <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-ceo-dario-amodei-ai-generated-code">Anthropic CEO Dario Amodei's 2025 prediction</a> that AI could eliminate half of all entry-level white-collar jobs within five years. </p><p>"The public, workers, young people, and university students are watching the rapid development of AI with more fear than excitement, with real concern for what it will do to jobs, particularly at entry levels, and, therefore, the prospects for our young people and the economy in general, said Professor Bobby Duffy, director of the Policy Institute at King's College London.</p><p>Just one-in-five members of the public believe the education system is preparing young people well for a world shaped by AI. And public appetite for government intervention is strong, with two-thirds saying AI companies should be closely regulated, even if that slows development. </p><p>And, across all groups, more than half supported both government-guaranteed retraining for workers <a href="https://www.itpro.com/machine-learning/34322/replaced-by-an-ai-would-you-retrain-for-a-new-job">displaced by AI</a> and a tax on companies that replace workers with AI to fund retraining.</p><p>"This survey gives a really interesting window into how British students, workers, and employers feel about AI. Some of the main concerns held by the public, such as fewer job openings, a contraction in entry-level roles, and increased pressure on white-collar work, echo what I find in my own research on AI and the UK labour market," said Dr Bouke Klein Teeselink, lecturer in philosophy, politics and economics at KCL.</p><p>"But none of these effects is fixed. With the right training, policies, and institutional support, there is a clear path forward to a more hopeful future, with rising productivity, broader opportunity, higher incomes, and faster scientific progress."</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/british-public-deeply-fearful-of-ai-with-one-in-five-even-thinking-it-will-lead-to-civil-unrest</link>
                                                                            <description>
                            <![CDATA[ Research by King's College London suggests people think AI's impact will be worse than a normal recession ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">gaSo3v9XtspF6CVMWJgkUj</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/ZgD62SYPvrCzZZCaK5frhC-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 19 May 2026 13:30:56 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:05 +0000</updated>
                                                                                                                                            <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/ZgD62SYPvrCzZZCaK5frhC-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[A photo of a messy, modern shared office, with monitors, wires, keyboards, and papers on a desk against an exposed brick background.]]></media:description>                                                            <media:text><![CDATA[A photo of a messy, modern shared office, with monitors, wires, keyboards, and papers on a desk against an exposed brick background.]]></media:text>
                                <media:title type="plain"><![CDATA[A photo of a messy, modern shared office, with monitors, wires, keyboards, and papers on a desk against an exposed brick background.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/ZgD62SYPvrCzZZCaK5frhC-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>The British public is deeply fearful of AI, with one-in-five even thinking it will lead to civil unrest.</p><p>New research by King's College London's Institute for Artificial Intelligence and Policy Institute has found that seven-in-ten people are worried about the economic impacts of AI, six-in-ten think it will eliminate more jobs than it creates, and half think its impact will be worse than a normal recession.</p><p>Almost half the public – 48% – said they'd rather avoid AI, while 41% are afraid of it. Only 24% think it's positive for humanity, compared with 39% who disagree.</p><p>There's a noticeable gender gap here, with women more likely to disagree than agree that AI is positive for humanity, at 44% versus 18%; positive for the UK (34% vs 23%; and that AI will improve their lives (41% vs 22%). </p><p>University students are more optimistic than the wider public, with 47% saying AI is positive for the UK and 43% that it is positive for humanity, compared with just 28% and 24% of the general public, respectively.</p><p>Meanwhile, seven-in-ten workers are worried about the economic impact of <a href="https://www.itpro.com/business/business-strategy/ai-job-losses-great-recession-us-forrester">AI job losses</a> – and the figure's almost as high for employers, at 64%. Most of the public (57%) thinks AI will lead to widespread unemployment.</p><p>People are particularly worried about the impacts on entry-level jobs and young people more generally, with nearly six-in-ten agreeing with <a href="https://www.itpro.com/technology/artificial-intelligence/anthropic-ceo-dario-amodei-ai-generated-code">Anthropic CEO Dario Amodei's 2025 prediction</a> that AI could eliminate half of all entry-level white-collar jobs within five years. </p><p>"The public, workers, young people, and university students are watching the rapid development of AI with more fear than excitement, with real concern for what it will do to jobs, particularly at entry levels, and, therefore, the prospects for our young people and the economy in general, said Professor Bobby Duffy, director of the Policy Institute at King's College London.</p><p>Just one-in-five members of the public believe the education system is preparing young people well for a world shaped by AI. And public appetite for government intervention is strong, with two-thirds saying AI companies should be closely regulated, even if that slows development. </p><p>And, across all groups, more than half supported both government-guaranteed retraining for workers <a href="https://www.itpro.com/machine-learning/34322/replaced-by-an-ai-would-you-retrain-for-a-new-job">displaced by AI</a> and a tax on companies that replace workers with AI to fund retraining.</p><p>"This survey gives a really interesting window into how British students, workers, and employers feel about AI. Some of the main concerns held by the public, such as fewer job openings, a contraction in entry-level roles, and increased pressure on white-collar work, echo what I find in my own research on AI and the UK labour market," said Dr Bouke Klein Teeselink, lecturer in philosophy, politics and economics at KCL.</p><p>"But none of these effects is fixed. With the right training, policies, and institutional support, there is a clear path forward to a more hopeful future, with rising productivity, broader opportunity, higher incomes, and faster scientific progress."</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Anthropic lets Glasswing partners publicly share Mythos flaws  ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Anthropic will allow companies using Mythos to share what they find publicly, contrary to initial confidentiality agreements with partners — meaning any flaws spotted in the AI model can be widely published. </p><p>Last month, <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity"><u>Anthropic unveiled Project Glasswing</u></a>, a partnership that brings together Apple, AWS, CrowdStrike, Google, Microsoft, and more in an effort to secure software from the impacts of AI. </p><p>That was sparked by Anthropic's Claude Mythos model, which the company claims has found thousands of serious flaws in every major operating system and web browser. Because of that, Anthropic is keeping Mythos under lock and key and offering access only to trusted partners — a move designed to keep the technology out of the hands of hackers, though some have <a href="https://www.youtube.com/watch?v=PsKVSHjres4"><u>suggested</u></a> that caution is at least in part a marketing move. </p><p>Initially, Anthropic said it would share any findings from Glasswing partners using Mythos, suggesting that the other companies would not be allowed to publish any flaws or concerns they discovered. </p><p>Now, Anthropic has confirmed that its partners will indeed be allowed to share any threats they discover with each other and external companies, as well as government authorities, the public, and the media, according to a report from <a href="https://www.reuters.com/technology/anthropic-let-partners-share-mythos-cybersecurity-findings-with-others-2026-05-18/"><u>Reuters</u></a>. </p><p>"While there was never a specific Glasswing NDA, confidentiality protections ‌were ⁠something partners asked for at the outset and were built into agreements partners signed," an ​Anthropic spokesperson told the news agency via a statement.</p><p>"As the program ⁠has matured, we've adapted them to ensure key information can be shared broadly — including outside the program — for maximum defensive ​impact," the statement added. </p><p>Allowing Glasswing partners to work with affected companies with Mythos-spotted flaws allows the idea of responsible disclosure to continue, as any issues can be shared and solved — without necessarily keeping Anthropic at the centre of such efforts. That could have risked a bottleneck, especially if the rate of flaw discovery continues to be high. </p><h2 id="ai-vs-security">AI vs security</h2><p>More generally, Anthropic is keeping Mythos Preview locked down over fears of its wider impact. "For cyber defenders to come out ahead, we need to act now," it <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity"><u>said</u></a> at the time.  </p><p>OpenAI clearly agrees, and last week unveiled its own version of Glasswing, called Daybreak. Built on its Codex Security tool and GPT 5.5, OpenAI aims to build AI-powered protections into software, rather than just using the technology to spot flaws — though, of course, there will still be plenty of that. </p><p>"AI can now help defenders reason across codebases, identify subtle vulnerabilities, validate fixes, analyze unfamiliar systems, and move from discovery to remediation faster," OpenAI <a href="https://openai.com/daybreak/"><u>said in a blog post</u></a> at the time. "Because those same capabilities can be misused, Daybreak pairs expanded defensive capability with trust, verification, proportional safeguards, and accountability."</p><p>In a briefing, analyst firm Gartner warned that the time between patch and exploit has now collapsed, impacting how security works across all industries — <a href="https://www.itpro.com/security/ai-is-raising-the-stakes-for-cyber-professionals-claude-mythos-just-took-things-to-another-level"><u>AI is raising the stakes for all cyber professionals. </u></a></p><p>"Security providers should embrace adoption of such AI tools to assess all code in the wild and code in development to rapidly address discovered exploits," Gartner noted, adding: "In addition, as consumers of software with security implications, security providers should accelerate their exposure to threats from integrated code through proactive remediation coordination with partners." </p><p>In other words, work with AI developers, or risk being left behind. </p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/anthropic-lets-glasswing-partners-publicly-share-mythos-flaws</link>
                                                                            <description>
                            <![CDATA[ Project Glasswing participants will be able to go public with what they find using Mythos Preview ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">VQ8H2zA99PHFYnJkPsEeeU</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/B5sequrZU2Sim7e6k6gTnH-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 19 May 2026 12:59:40 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:12 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/B5sequrZU2Sim7e6k6gTnH-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[The Anthropic logo, in white on a dark background lit by a web-like light.]]></media:description>                                                            <media:text><![CDATA[The Anthropic logo, in white on a dark background lit by a web-like light.]]></media:text>
                                <media:title type="plain"><![CDATA[The Anthropic logo, in white on a dark background lit by a web-like light.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/B5sequrZU2Sim7e6k6gTnH-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Anthropic will allow companies using Mythos to share what they find publicly, contrary to initial confidentiality agreements with partners — meaning any flaws spotted in the AI model can be widely published. </p><p>Last month, <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity"><u>Anthropic unveiled Project Glasswing</u></a>, a partnership that brings together Apple, AWS, CrowdStrike, Google, Microsoft, and more in an effort to secure software from the impacts of AI. </p><p>That was sparked by Anthropic's Claude Mythos model, which the company claims has found thousands of serious flaws in every major operating system and web browser. Because of that, Anthropic is keeping Mythos under lock and key and offering access only to trusted partners — a move designed to keep the technology out of the hands of hackers, though some have <a href="https://www.youtube.com/watch?v=PsKVSHjres4"><u>suggested</u></a> that caution is at least in part a marketing move. </p><p>Initially, Anthropic said it would share any findings from Glasswing partners using Mythos, suggesting that the other companies would not be allowed to publish any flaws or concerns they discovered. </p><p>Now, Anthropic has confirmed that its partners will indeed be allowed to share any threats they discover with each other and external companies, as well as government authorities, the public, and the media, according to a report from <a href="https://www.reuters.com/technology/anthropic-let-partners-share-mythos-cybersecurity-findings-with-others-2026-05-18/"><u>Reuters</u></a>. </p><p>"While there was never a specific Glasswing NDA, confidentiality protections ‌were ⁠something partners asked for at the outset and were built into agreements partners signed," an ​Anthropic spokesperson told the news agency via a statement.</p><p>"As the program ⁠has matured, we've adapted them to ensure key information can be shared broadly — including outside the program — for maximum defensive ​impact," the statement added. </p><p>Allowing Glasswing partners to work with affected companies with Mythos-spotted flaws allows the idea of responsible disclosure to continue, as any issues can be shared and solved — without necessarily keeping Anthropic at the centre of such efforts. That could have risked a bottleneck, especially if the rate of flaw discovery continues to be high. </p><h2 id="ai-vs-security">AI vs security</h2><p>More generally, Anthropic is keeping Mythos Preview locked down over fears of its wider impact. "For cyber defenders to come out ahead, we need to act now," it <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity"><u>said</u></a> at the time.  </p><p>OpenAI clearly agrees, and last week unveiled its own version of Glasswing, called Daybreak. Built on its Codex Security tool and GPT 5.5, OpenAI aims to build AI-powered protections into software, rather than just using the technology to spot flaws — though, of course, there will still be plenty of that. </p><p>"AI can now help defenders reason across codebases, identify subtle vulnerabilities, validate fixes, analyze unfamiliar systems, and move from discovery to remediation faster," OpenAI <a href="https://openai.com/daybreak/"><u>said in a blog post</u></a> at the time. "Because those same capabilities can be misused, Daybreak pairs expanded defensive capability with trust, verification, proportional safeguards, and accountability."</p><p>In a briefing, analyst firm Gartner warned that the time between patch and exploit has now collapsed, impacting how security works across all industries — <a href="https://www.itpro.com/security/ai-is-raising-the-stakes-for-cyber-professionals-claude-mythos-just-took-things-to-another-level"><u>AI is raising the stakes for all cyber professionals. </u></a></p><p>"Security providers should embrace adoption of such AI tools to assess all code in the wild and code in development to rapidly address discovered exploits," Gartner noted, adding: "In addition, as consumers of software with security implications, security providers should accelerate their exposure to threats from integrated code through proactive remediation coordination with partners." </p><p>In other words, work with AI developers, or risk being left behind. </p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Pope Leo launches AI commission  ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Pope Leo has weighed in again on AI, with the creation of a study group on <a href="https://www.itpro.com/strategy/29848/is-artificial-intelligence-safe">artificial intelligence</a>.</p><p>The Vatican <a href="https://press.vatican.va/content/salastampa/en/bollettino/pubblico/2026/05/16/260516b.html">said</a> that the Inter-Dicasterial Commission on Artificial Intelligence is being set up in light of the acceleration in the use of AI over recent years. It will, it said, look at AI's potential effects on human beings and on humanity as a whole.</p><p>It will be made up of representatives from the Dicastery for Promoting Integral Human Development, the Dicastery for the Doctrine of the Faith, the Dicastery for Culture and Education, the Dicastery for Communication, the Pontifical Academy for Life, the Pontifical Academy of Sciences, and the Pontifical Academy of Social Sciences.</p><p>During the first year, it will be coordinated by the Dicastery for Promoting Integral Human Development, with the role then shifting to another participating institution, again for one year.</p><p>Earlier this year, the Pope issued a <a href="https://www.vatican.va/content/leo-xiv/en/messages/communications/documents/20260124-messaggio-comunicazioni-sociali.html">message</a> for World Communications Day, expressing serious concerns about the use of AI. <a href="https://www.itpro.com/data-insights/30212/what-is-an-algorithm">Algorithms</a> designed to maximize engagement on social media, he warned, reduce people's ability to listen and think critically, increasing social polarization.</p><p>"Although AI can provide support and assistance in managing tasks related to communication, in the long run, choosing to evade the effort of thinking for ourselves and settling for artificial statistical compilations threatens to diminish our cognitive, emotional, and communication skills," he said.</p><p>"In recent years, artificial intelligence systems have increasingly taken control of the production of texts, music, and videos. This puts much of the human creative industry at risk of being dismantled and replaced with the label 'Powered by AI', turning people into passive consumers of unthought thoughts and anonymous products without ownership or love."</p><p>He called on online platforms to ensure that their business strategies aren't guided solely by the profit motive, but to consider the common good. The creators and developers of AI models should practice transparency and social responsibility, he said, while national legislators and supranational regulators should ensure respect for human dignity.</p><p>"All stakeholders – from the tech industry to legislators, from creative companies to academia, from artists to journalists and educators – must be involved in building and implementing informed and responsible digital citizenship," he said.</p><p>In the coming days, Pope Leo is expected to release a new encyclical, designed to provide guidance to bishops on moral questions – and widely expected to repeat his <a href="https://www.itpro.com/technology/artificial-intelligence/concerns-are-mounting-over-the-cognitive-impact-of-ai-as-workers-report-experiencing-brain-fry-and-its-causing-increased-employee-errors-decision-fatigue-and-intention-to-quit">concerns on AI</a>.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/pope-leo-launches-ai-commission-over-concerns-that-people-will-turn-into-passive-consumers-of-unthought-thoughts-and-anonymous-products-without-ownership-or-love</link>
                                                                            <description>
                            <![CDATA[ With a new encyclical expected to focus on the technology, the Pope is concerned about its effects on human dignity ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">ifdY8DzDRsMkXt5LCBUwVC</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/q3MoWrLLLCQMoWvSx2noRB-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Mon, 18 May 2026 10:15:46 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:34 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/q3MoWrLLLCQMoWvSx2noRB-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Pope Leo]]></media:description>                                                            <media:text><![CDATA[Pope Leo]]></media:text>
                                <media:title type="plain"><![CDATA[Pope Leo]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/q3MoWrLLLCQMoWvSx2noRB-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Pope Leo has weighed in again on AI, with the creation of a study group on <a href="https://www.itpro.com/strategy/29848/is-artificial-intelligence-safe">artificial intelligence</a>.</p><p>The Vatican <a href="https://press.vatican.va/content/salastampa/en/bollettino/pubblico/2026/05/16/260516b.html">said</a> that the Inter-Dicasterial Commission on Artificial Intelligence is being set up in light of the acceleration in the use of AI over recent years. It will, it said, look at AI's potential effects on human beings and on humanity as a whole.</p><p>It will be made up of representatives from the Dicastery for Promoting Integral Human Development, the Dicastery for the Doctrine of the Faith, the Dicastery for Culture and Education, the Dicastery for Communication, the Pontifical Academy for Life, the Pontifical Academy of Sciences, and the Pontifical Academy of Social Sciences.</p><p>During the first year, it will be coordinated by the Dicastery for Promoting Integral Human Development, with the role then shifting to another participating institution, again for one year.</p><p>Earlier this year, the Pope issued a <a href="https://www.vatican.va/content/leo-xiv/en/messages/communications/documents/20260124-messaggio-comunicazioni-sociali.html">message</a> for World Communications Day, expressing serious concerns about the use of AI. <a href="https://www.itpro.com/data-insights/30212/what-is-an-algorithm">Algorithms</a> designed to maximize engagement on social media, he warned, reduce people's ability to listen and think critically, increasing social polarization.</p><p>"Although AI can provide support and assistance in managing tasks related to communication, in the long run, choosing to evade the effort of thinking for ourselves and settling for artificial statistical compilations threatens to diminish our cognitive, emotional, and communication skills," he said.</p><p>"In recent years, artificial intelligence systems have increasingly taken control of the production of texts, music, and videos. This puts much of the human creative industry at risk of being dismantled and replaced with the label 'Powered by AI', turning people into passive consumers of unthought thoughts and anonymous products without ownership or love."</p><p>He called on online platforms to ensure that their business strategies aren't guided solely by the profit motive, but to consider the common good. The creators and developers of AI models should practice transparency and social responsibility, he said, while national legislators and supranational regulators should ensure respect for human dignity.</p><p>"All stakeholders – from the tech industry to legislators, from creative companies to academia, from artists to journalists and educators – must be involved in building and implementing informed and responsible digital citizenship," he said.</p><p>In the coming days, Pope Leo is expected to release a new encyclical, designed to provide guidance to bishops on moral questions – and widely expected to repeat his <a href="https://www.itpro.com/technology/artificial-intelligence/concerns-are-mounting-over-the-cognitive-impact-of-ai-as-workers-report-experiencing-brain-fry-and-its-causing-increased-employee-errors-decision-fatigue-and-intention-to-quit">concerns on AI</a>.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ AI is getting better at security ]]></title>
                                                                                                <dc:content><![CDATA[ <p>AI models are getting better at handling more complex security tasks, doubling results in one benchmark in the last few months – and that's before the arrival of security-focused models, notably Anthropic's <a href="https://www.itpro.com/security/ai-is-raising-the-stakes-for-cyber-professionals-claude-mythos-just-took-things-to-another-level">Claude Mythos</a> and OpenAI's GPT-5.5.</p><p>That's according to the UK's AI Security Institute (AISI), which tracks the potential impact of AI on the security industry and efforts to protect organisations, and found newer models had doubled the length of cyber tasks they could complete in just 4.7 months – much faster than expected. </p><p>"In February 2026, we internally estimated that the length of cyber tasks AI models could complete had doubled every 4.7 months since late 2024 – already an acceleration from our November 2025 estimate of 8 months," the organisation said in a blog post. "Since then, AISI reported on two new models, Claude Mythos Preview and GPT-5.5, which substantially exceeded both doubling rate trends."</p><p>AISI added: "It is unclear whether this represents a new, faster trend."</p><p>That follows the release of Claude Mythos, which sparked concerns that companies wouldn't be able to keep up with AI security, as well as GPT-5.5 Cyber last week, with <a href="https://openai.com/index/gpt-5-5-with-trusted-access-for-cyber/">OpenAI releasing</a> the security focused model in a limited preview with access only to security professionals, amid fears that generative AI was accelerating a <a href="https://www.itpro.com/security/generative-ai-attacks-are-accelerating-at-an-alarming-rate">security arms race</a>. Indeed, Forescout VP of security intelligence Rik Ferguson last week said AI tools are now "<a href="https://www.itpro.com/security/ai-is-now-a-standard-part-of-the-attacker-toolkit">a standard part of the attacker toolkit</a>."</p><h2 id="how-the-aisa-tests">How the AISA tests </h2><p>These results are based on a time-horizon benchmark, which tracks the success rate of AI models on tasks of different lengths based on how long a human expert would take on the same task. For example, one set of tests includes reverse engineering and web exploits in self-contained setups. AISA is looking for a model to succeed 80% of the time to be considered capable of doing a task of a certain length. </p><p>AISI admits the time horizon benchmark is imperfect. "They are inexact predictors of performance; AI struggles with some tasks humans do quickly, and easily completes others that humans find hard," the blog post noted. "However, we use this type of benchmark because it offers a measure of AI autonomy from which we can draw trends." </p><p>Plus, the tests only include some capabilities that would be necessary to run a real-world attack. Alongside that, AISI limits the models to 2.5m tokens to maintain comparability across results. </p><p>But AISI said that the 2.5m token cap limits the success of models, as without that cap, the "success rates are so high that time horizons become impossible to calculate." But the organisation also added that its own tests are now too short, meaning it's not possible to reveal at what point model reliability would start to fail on a longer task; the longest task is 12 hours. </p><p>"No single benchmark result should be read as a precise measure of AI capability," the post noted, adding: "Regardless, the direction of change and rapid growth have been consistent across the models, methodological choices, and independent data we examined."</p><p>New evaluation methods were in development, the AISI added. </p><h2 id="what-this-means-for-security">What this means for security</h2><p>The AISI said it was unclear how AI's pace of progress would continue, or how the technology's capabilities would work against real-world systems. But the agency said it was clear that AI was bringing opportunities and risks. </p><p>"The time to invest in strong security baselines is now," the AISI post warned. "Frontier AI can strengthen attackers as well as defenders, and there is a critical window to build resilience."</p><p>That was echoed by Palo Alto Networks this week, with CTO Lee Klarich warning that AI cyberattacks would become the new normal in the next few months. "This impending vulnerability deluge demands urgency," he <a href="https://www.paloaltonetworks.com/blog/2026/05/defenders-guide-frontier-ai-impact-cybersecurity-may-2026-update/">wrote</a>. </p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/ai-is-getting-better-at-security-and-its-doing-it-faster-than-expected</link>
                                                                            <description>
                            <![CDATA[ UK AISI warns that AI models are already exceeding existing benchmarks for testing ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">C9j6iQdEoYMD8kbYJm56Rj</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/Da8G5KZK7gmoYomN8iKNVZ-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 14 May 2026 12:00:02 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:13 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/Da8G5KZK7gmoYomN8iKNVZ-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[A depiction of AI software ]]></media:description>                                                            <media:text><![CDATA[A depiction of AI software ]]></media:text>
                                <media:title type="plain"><![CDATA[A depiction of AI software ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/Da8G5KZK7gmoYomN8iKNVZ-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>AI models are getting better at handling more complex security tasks, doubling results in one benchmark in the last few months – and that's before the arrival of security-focused models, notably Anthropic's <a href="https://www.itpro.com/security/ai-is-raising-the-stakes-for-cyber-professionals-claude-mythos-just-took-things-to-another-level">Claude Mythos</a> and OpenAI's GPT-5.5.</p><p>That's according to the UK's AI Security Institute (AISI), which tracks the potential impact of AI on the security industry and efforts to protect organisations, and found newer models had doubled the length of cyber tasks they could complete in just 4.7 months – much faster than expected. </p><p>"In February 2026, we internally estimated that the length of cyber tasks AI models could complete had doubled every 4.7 months since late 2024 – already an acceleration from our November 2025 estimate of 8 months," the organisation said in a blog post. "Since then, AISI reported on two new models, Claude Mythos Preview and GPT-5.5, which substantially exceeded both doubling rate trends."</p><p>AISI added: "It is unclear whether this represents a new, faster trend."</p><p>That follows the release of Claude Mythos, which sparked concerns that companies wouldn't be able to keep up with AI security, as well as GPT-5.5 Cyber last week, with <a href="https://openai.com/index/gpt-5-5-with-trusted-access-for-cyber/">OpenAI releasing</a> the security focused model in a limited preview with access only to security professionals, amid fears that generative AI was accelerating a <a href="https://www.itpro.com/security/generative-ai-attacks-are-accelerating-at-an-alarming-rate">security arms race</a>. Indeed, Forescout VP of security intelligence Rik Ferguson last week said AI tools are now "<a href="https://www.itpro.com/security/ai-is-now-a-standard-part-of-the-attacker-toolkit">a standard part of the attacker toolkit</a>."</p><h2 id="how-the-aisa-tests">How the AISA tests </h2><p>These results are based on a time-horizon benchmark, which tracks the success rate of AI models on tasks of different lengths based on how long a human expert would take on the same task. For example, one set of tests includes reverse engineering and web exploits in self-contained setups. AISA is looking for a model to succeed 80% of the time to be considered capable of doing a task of a certain length. </p><p>AISI admits the time horizon benchmark is imperfect. "They are inexact predictors of performance; AI struggles with some tasks humans do quickly, and easily completes others that humans find hard," the blog post noted. "However, we use this type of benchmark because it offers a measure of AI autonomy from which we can draw trends." </p><p>Plus, the tests only include some capabilities that would be necessary to run a real-world attack. Alongside that, AISI limits the models to 2.5m tokens to maintain comparability across results. </p><p>But AISI said that the 2.5m token cap limits the success of models, as without that cap, the "success rates are so high that time horizons become impossible to calculate." But the organisation also added that its own tests are now too short, meaning it's not possible to reveal at what point model reliability would start to fail on a longer task; the longest task is 12 hours. </p><p>"No single benchmark result should be read as a precise measure of AI capability," the post noted, adding: "Regardless, the direction of change and rapid growth have been consistent across the models, methodological choices, and independent data we examined."</p><p>New evaluation methods were in development, the AISI added. </p><h2 id="what-this-means-for-security">What this means for security</h2><p>The AISI said it was unclear how AI's pace of progress would continue, or how the technology's capabilities would work against real-world systems. But the agency said it was clear that AI was bringing opportunities and risks. </p><p>"The time to invest in strong security baselines is now," the AISI post warned. "Frontier AI can strengthen attackers as well as defenders, and there is a critical window to build resilience."</p><p>That was echoed by Palo Alto Networks this week, with CTO Lee Klarich warning that AI cyberattacks would become the new normal in the next few months. "This impending vulnerability deluge demands urgency," he <a href="https://www.paloaltonetworks.com/blog/2026/05/defenders-guide-frontier-ai-impact-cybersecurity-may-2026-update/">wrote</a>. </p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Changes to EU AI Act implementation deadlines welcomed by industry ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Changes made to the <a href="https://www.itpro.com/business/policy-and-legislation/how-the-eu-ai-act-compares-to-other-international-regulatory-approaches">EU AI Act</a> last week have been welcomed by industry stakeholders, but the move doesn’t mean enterprises can take their eye off the ball on AI governance. </p><p>Last week, a <a href="https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/" target="_blank">provisional agreement</a> was reached between the European Parliament and EU Council on the Omnibus VII legislative package. This includes proposals which aim to simplify the union’s legislative landscape. </p><p>Under the deal, significant changes to the EU AI Act will be implemented, including a revised compliance timeline for AI systems deemed “high-risk” under the legislation, as well as extended exemptions for small-to-medium sized enterprises. </p><p>As part of the delayed high-risk rules, new application dates will be 2 December 2027 for stand-alone AI systems, and 2 August 2028 for high-risk AI systems embedded in products, according to the EU Council. </p><p>Similarly, agreement reinstates provider obligations to register AI systems in the EU database for high-risk systems. </p><p>Elsewhere, the provisional agreement also postpones the deadline for the implementation of AI regulatory sandboxes at a national level until 2 August 2027.</p><p>Henna Virkkunen, executive vice president for tech sovereignty, security, and democracy at the European Commission, said the changes will help streamline implementation of the legislation. </p><p>“Our businesses and citizens want two things from <a href="https://www.itpro.com/business/policy-and-legislation/the-second-enforcement-deadline-for-the-eu-ai-act-is-approaching-heres-what-businesses-need-to-know-about-the-general-purpose-ai-code-of-practice">AI rules</a>. They want to be able to innovate and feel safe,” she said. </p><p>“Today’s agreement does both. With simpler and innovation-friendly rules, we make it easier to innovate without lowering the bar on safety. We are also making sure the tools supporting EU companies for a smooth implementation of the AI Act are ready. </p><h2 id="meaningful-changes-to-the-eu-ai-act">“Meaningful changes” to the EU AI Act</h2><p>The changes made to the legislation have been welcomed by industry stakeholders as a positive step toward streamlined implementation. </p><p>Mark Weir, regional director for the UK &amp; Ireland at Check Point Software, said the alterations represent a “meaningful evolution in EU <a href="https://www.itpro.com/technology/artificial-intelligence/organizations-face-ticking-timebomb-over-ai-governance">AI governance</a>” that balances stronger protections alongside a practical approach to compliance. </p><p>“By harmonizing implementation across member states and reducing overlapping administrative obligations, the legislation removes genuine friction for organizations operating across Europe,” he said. </p><p>“Without clear guidance, even well-intentioned organisations struggle to translate broad principles into consistent practice,” Weir added. </p><p>“This closes a gap that has caused real uncertainty in the market. Taken together, these changes point toward a regulatory framework that is not just ambitious, but workable."</p><p>Weir noted that provider registration requirements are also a welcome move, and one that will give European legislators “real enforceability” over the EU’s digital sovereignty plans. </p><p>Levent Ergin, chief strategist for <a href="https://www.itpro.com/security/five-eyes-agencies-sound-alarm-over-risky-agentic-ai-deployments">agentic AI</a>, regulatory compliance, and sustainability at Informatica, echoed Weir’s comments but noted that the changes “shouldn’t detract from the need for better AI governance”. </p><p>“Businesses still need to ensure the data feeding their AI systems is governed, explainable, and built on trusted data foundations,” he said. </p><p>“Compliance is only part of the enabler of safe AI. Stronger governance, human oversight, and trusted context will ultimately determine which businesses can scale AI confidently and which struggle to move beyond experimentation.”</p><p>Looking ahead, the European Parliament and EU Council must not formally adopt the agreement. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/business/policy-and-legislation/this-closes-a-gap-that-has-caused-real-uncertainty-in-the-market-changes-to-eu-ai-act-implementation-deadlines-welcomed-by-industry</link>
                                                                            <description>
                            <![CDATA[ New implementation deadlines for the EU AI Act could help remove “genuine friction” for European companies ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">wTrcGJSwGQ4XAiubLrguKi</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/YPLAJxoRSrPttgxeZWQkeG-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Mon, 11 May 2026 14:23:01 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:56 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Policy and Legislation]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                    <category><![CDATA[Business]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/YPLAJxoRSrPttgxeZWQkeG-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[European Union (EU) concept image showing flag on a digitized background with ripples flowing out from 12 stars.]]></media:description>                                                            <media:text><![CDATA[European Union (EU) concept image showing flag on a digitized background with ripples flowing out from 12 stars.]]></media:text>
                                <media:title type="plain"><![CDATA[European Union (EU) concept image showing flag on a digitized background with ripples flowing out from 12 stars.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/YPLAJxoRSrPttgxeZWQkeG-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Changes made to the <a href="https://www.itpro.com/business/policy-and-legislation/how-the-eu-ai-act-compares-to-other-international-regulatory-approaches">EU AI Act</a> last week have been welcomed by industry stakeholders, but the move doesn’t mean enterprises can take their eye off the ball on AI governance. </p><p>Last week, a <a href="https://www.consilium.europa.eu/en/press/press-releases/2026/05/07/artificial-intelligence-council-and-parliament-agree-to-simplify-and-streamline-rules/" target="_blank">provisional agreement</a> was reached between the European Parliament and EU Council on the Omnibus VII legislative package. This includes proposals which aim to simplify the union’s legislative landscape. </p><p>Under the deal, significant changes to the EU AI Act will be implemented, including a revised compliance timeline for AI systems deemed “high-risk” under the legislation, as well as extended exemptions for small-to-medium sized enterprises. </p><p>As part of the delayed high-risk rules, new application dates will be 2 December 2027 for stand-alone AI systems, and 2 August 2028 for high-risk AI systems embedded in products, according to the EU Council. </p><p>Similarly, agreement reinstates provider obligations to register AI systems in the EU database for high-risk systems. </p><p>Elsewhere, the provisional agreement also postpones the deadline for the implementation of AI regulatory sandboxes at a national level until 2 August 2027.</p><p>Henna Virkkunen, executive vice president for tech sovereignty, security, and democracy at the European Commission, said the changes will help streamline implementation of the legislation. </p><p>“Our businesses and citizens want two things from <a href="https://www.itpro.com/business/policy-and-legislation/the-second-enforcement-deadline-for-the-eu-ai-act-is-approaching-heres-what-businesses-need-to-know-about-the-general-purpose-ai-code-of-practice">AI rules</a>. They want to be able to innovate and feel safe,” she said. </p><p>“Today’s agreement does both. With simpler and innovation-friendly rules, we make it easier to innovate without lowering the bar on safety. We are also making sure the tools supporting EU companies for a smooth implementation of the AI Act are ready. </p><h2 id="meaningful-changes-to-the-eu-ai-act">“Meaningful changes” to the EU AI Act</h2><p>The changes made to the legislation have been welcomed by industry stakeholders as a positive step toward streamlined implementation. </p><p>Mark Weir, regional director for the UK &amp; Ireland at Check Point Software, said the alterations represent a “meaningful evolution in EU <a href="https://www.itpro.com/technology/artificial-intelligence/organizations-face-ticking-timebomb-over-ai-governance">AI governance</a>” that balances stronger protections alongside a practical approach to compliance. </p><p>“By harmonizing implementation across member states and reducing overlapping administrative obligations, the legislation removes genuine friction for organizations operating across Europe,” he said. </p><p>“Without clear guidance, even well-intentioned organisations struggle to translate broad principles into consistent practice,” Weir added. </p><p>“This closes a gap that has caused real uncertainty in the market. Taken together, these changes point toward a regulatory framework that is not just ambitious, but workable."</p><p>Weir noted that provider registration requirements are also a welcome move, and one that will give European legislators “real enforceability” over the EU’s digital sovereignty plans. </p><p>Levent Ergin, chief strategist for <a href="https://www.itpro.com/security/five-eyes-agencies-sound-alarm-over-risky-agentic-ai-deployments">agentic AI</a>, regulatory compliance, and sustainability at Informatica, echoed Weir’s comments but noted that the changes “shouldn’t detract from the need for better AI governance”. </p><p>“Businesses still need to ensure the data feeding their AI systems is governed, explainable, and built on trusted data foundations,” he said. </p><p>“Compliance is only part of the enabler of safe AI. Stronger governance, human oversight, and trusted context will ultimately determine which businesses can scale AI confidently and which struggle to move beyond experimentation.”</p><p>Looking ahead, the European Parliament and EU Council must not formally adopt the agreement. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Microsoft joins competitors in handing over AI models for advanced testing ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Microsoft, Google, and xAI have agreed to hand over their AI tools to the US Center for AI Standards and Innovation (CAISI) and the UK's AI Security Institute (AISI) for pre-deployment testing.</p><p>They will evaluate the firms' frontier models, assess safeguards, and help mitigate national security and large-scale public safety risks, Microsoft said. </p><p>"Well-constructed tests help us understand whether our systems are working as intended and delivering the benefits they are designed to provide. Testing also helps us stay ahead of risks, such as <a href="https://www.itpro.com/security/foreign-states-ramp-up-cyber-attacks-on-eu-with-ai-driven-phishing-and-ddos-campaigns">AI-driven cyber attacks</a> and other criminal misuses of AI systems, that can emerge once advanced AI systems are deployed in the world," said Natasha Crampton, Microsoft’s chief responsible AI officer.</p><p>"While Microsoft regularly undertakes many types of AI testing on its own, testing for national security and large-scale public safety risks necessarily must be a collaborative endeavor with governments. This type of testing depends on deep technical, scientific, and national security expertise that is uniquely held by institutions like CAISI in the US and AISI in the UK and the government agencies they work with." </p><p>In the US, Microsoft and the National Institute of Standards and Technology (NIST) will collaborate with CAISI on improving methodologies for adversarial assessments. </p><p>It will mean testing AI systems by examining unexpected behaviors, misuse pathways, and failure modes. This includes developing more systematic and reproducible approaches to evaluation, including shared frameworks, datasets, and workflows for assessing safety, security, and robustness risks in advanced AI systems.</p><p>“Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” said CAISI director Chris Fall. “These expanded industry collaborations help us scale our work in the public interest at a critical moment.”</p><h2 id="microsoft-strikes-agreement-with-uk-researchers">Microsoft strikes agreement with UK researchers</h2><p>In the UK, Microsoft will collaborate with AISI on research related to frontier safety and security, including ways of evaluating high-risk capabilities and the effectiveness of the safeguards used to address them</p><p>"The partnership will also include research into societal resilience, examining how conversational AI systems interact with users insensitive contexts," said AISI.</p><p>"As AI systems become increasingly capable, sustained two-way collaboration between government and companies developing and deploying <a href="https://www.itpro.com/technology/artificial-intelligence/aws-says-anyone-can-build-an-ai-model-with-amazon-nova-forge">frontier AI</a> is essential to advance our joint understanding of large-scale risks to public safety and national security."</p><p>Microsoft said future plans include collaborating with other AI institutes around the world, sharing priorities and methodologies for testing through the International Network for AI Measurement, Evaluation and Science. </p><p>The company is also working with Frontier Model Forum (FMF), an initiative dedicated to advancing the science and practice of frontier <a href="https://www.itpro.com/security/observability-will-be-key-to-agentic-ai-safety-says-microsoft-security-exec">AI safety and security</a>, to support independent research and promote transparency around risk mitigation strategies.  </p><p>It is also contributing to MLCommons, a multistakeholder non-profit that develops and operationalizes testing tools such as AILuminate, a family of safety and security benchmarks. </p><p>"As AI capabilities advance, so too must the rigor of the testing and safeguards that underpin them. We will apply what we learn from these partnerships directly into how we design, test, and deploy AI systems, ensuring that progress in evaluation science translates into safer, more secure products for our customers," said Crampton. </p><p>"As these partnerships progress, we will share what we learn and look for opportunities to apply insights and best practices to AI testing more broadly."</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/microsoft-joins-competitors-in-handing-over-ai-models-for-advanced-testing</link>
                                                                            <description>
                            <![CDATA[ US and UK government agencies will evaluate the firm's frontier models, along with those from Google and xAI ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">FFZSDHS7GHj4dS3DnXrWnW</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/Deb4XXCJdzDg8DCaViLow3-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 07 May 2026 10:30:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:07 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/Deb4XXCJdzDg8DCaViLow3-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Microsoft logo and branding pictured at Mobile World Congress (MWC) 2025 in Barcelona, Spain. ]]></media:description>                                                            <media:text><![CDATA[Microsoft logo and branding pictured at Mobile World Congress (MWC) 2025 in Barcelona, Spain. ]]></media:text>
                                <media:title type="plain"><![CDATA[Microsoft logo and branding pictured at Mobile World Congress (MWC) 2025 in Barcelona, Spain. ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/Deb4XXCJdzDg8DCaViLow3-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Microsoft, Google, and xAI have agreed to hand over their AI tools to the US Center for AI Standards and Innovation (CAISI) and the UK's AI Security Institute (AISI) for pre-deployment testing.</p><p>They will evaluate the firms' frontier models, assess safeguards, and help mitigate national security and large-scale public safety risks, Microsoft said. </p><p>"Well-constructed tests help us understand whether our systems are working as intended and delivering the benefits they are designed to provide. Testing also helps us stay ahead of risks, such as <a href="https://www.itpro.com/security/foreign-states-ramp-up-cyber-attacks-on-eu-with-ai-driven-phishing-and-ddos-campaigns">AI-driven cyber attacks</a> and other criminal misuses of AI systems, that can emerge once advanced AI systems are deployed in the world," said Natasha Crampton, Microsoft’s chief responsible AI officer.</p><p>"While Microsoft regularly undertakes many types of AI testing on its own, testing for national security and large-scale public safety risks necessarily must be a collaborative endeavor with governments. This type of testing depends on deep technical, scientific, and national security expertise that is uniquely held by institutions like CAISI in the US and AISI in the UK and the government agencies they work with." </p><p>In the US, Microsoft and the National Institute of Standards and Technology (NIST) will collaborate with CAISI on improving methodologies for adversarial assessments. </p><p>It will mean testing AI systems by examining unexpected behaviors, misuse pathways, and failure modes. This includes developing more systematic and reproducible approaches to evaluation, including shared frameworks, datasets, and workflows for assessing safety, security, and robustness risks in advanced AI systems.</p><p>“Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” said CAISI director Chris Fall. “These expanded industry collaborations help us scale our work in the public interest at a critical moment.”</p><h2 id="microsoft-strikes-agreement-with-uk-researchers">Microsoft strikes agreement with UK researchers</h2><p>In the UK, Microsoft will collaborate with AISI on research related to frontier safety and security, including ways of evaluating high-risk capabilities and the effectiveness of the safeguards used to address them</p><p>"The partnership will also include research into societal resilience, examining how conversational AI systems interact with users insensitive contexts," said AISI.</p><p>"As AI systems become increasingly capable, sustained two-way collaboration between government and companies developing and deploying <a href="https://www.itpro.com/technology/artificial-intelligence/aws-says-anyone-can-build-an-ai-model-with-amazon-nova-forge">frontier AI</a> is essential to advance our joint understanding of large-scale risks to public safety and national security."</p><p>Microsoft said future plans include collaborating with other AI institutes around the world, sharing priorities and methodologies for testing through the International Network for AI Measurement, Evaluation and Science. </p><p>The company is also working with Frontier Model Forum (FMF), an initiative dedicated to advancing the science and practice of frontier <a href="https://www.itpro.com/security/observability-will-be-key-to-agentic-ai-safety-says-microsoft-security-exec">AI safety and security</a>, to support independent research and promote transparency around risk mitigation strategies.  </p><p>It is also contributing to MLCommons, a multistakeholder non-profit that develops and operationalizes testing tools such as AILuminate, a family of safety and security benchmarks. </p><p>"As AI capabilities advance, so too must the rigor of the testing and safeguards that underpin them. We will apply what we learn from these partnerships directly into how we design, test, and deploy AI systems, ensuring that progress in evaluation science translates into safer, more secure products for our customers," said Crampton. </p><p>"As these partnerships progress, we will share what we learn and look for opportunities to apply insights and best practices to AI testing more broadly."</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ AI adoption is accelerating in the UK, but ‘trust is not keeping pace’ ]]></title>
                                                                                                <dc:content><![CDATA[ <p>UK consumers are growing increasingly concerned over AI-related controls, accountability, and governance, according to new research from EY. </p><p>The consultancy’s 2026 AI Sentiment Index found that nearly three-quarters (74%) of Brits have used AI recently, with the technology now deeply integrated into daily life and work. </p><p>Only 14% are comfortable with fully autonomous AI solutions, however, highlighting a growing demand for oversight. Just 43% of respondents said they trust companies using AI-powered services, while only 41% trust government agencies to manage AI-related data effectively. </p><p>It’s those who use AI most frequently, particularly educated, white‑collar workers, that are most concerned.</p><p>“AI adoption in the UK is rapidly advancing, but trust is not keeping pace with technological capability. Whilst consumers are engaging with AI every day, many still want greater clarity about who is accountable when decisions are made on their behalf," said Matthew Ringelheim, EY UK and Ireland AI leader.</p><p>“This is a critical moment for organizations. As AI systems become more autonomous, trust must be embedded through strong data foundations, clear accountability and visible human oversight. Our research shows UK users want greater control and transparency, reinforcing the need to move beyond AI adoption for its own sake."</p><h2 id="ai-benefits-are-becoming-clear">AI benefits are becoming clear</h2><p>While 73% of respondents are concerned about AI systems being hacked or breached, they're happy to engage with AI when they see clear benefits. </p><p>This tends to mean using AI in practical, outcome‑driven contexts, particularly where systems can improve response times (59%) and reliability (35%), or create better value for money (52%).</p><p>In more emotionally sensitive or high stakes situations, EY found users seem more comfortable using AI where it’s clear that safeguards, privacy, and consistent decision-making are in place. </p><p>Half of all UK respondents, for example, said they had consciously used health or wellness applications in the last six months, while 35% used it for financial activities where consistency and data protection are critical.</p><p>The message for businesses is that encouraging more use of AI will depend on reassuring users that strong governance measures are in place. </p><h2 id="skills-and-awareness-are-key">Skills and awareness are key</h2><p>Elsewhere in the survey, EY noted that UK users reported a lack of appropriate skills and training in the technology. Just 23% of respondents said they had received significant <a href="https://www.itpro.com/technology/artificial-intelligence/aws-to-give-ai-skills-to-100-000-people-in-the-uk-by-2030">training or education on AI</a> in the workplace. </p><p>Ringelheim said skills development plays a “critical role” in both bolstering adoption rates, but also building trust in the systems used on a daily basis. </p><p>“As <a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> become more capable, people will need greater confidence in how they’re used at work and clearer, practical guidance on how to use them responsibly," he commented. </p><p>"Training also better equips users to better spot errors, challenge outputs and make more informed decisions on when to rely on AI and when to escalate human judgement. Workforce confidence — built through the right skills — will be decisive in turning AI momentum into long‑term growth for the UK.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/ai-adoption-is-accelerating-in-the-uk-but-trust-is-not-keeping-pace</link>
                                                                            <description>
                            <![CDATA[ Organizations need to do more to reassure customers over governance ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">QuC6NGJpBE5SXWK2sp93bL</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/PbaraXgyBf5cecgbD4TJLU-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Wed, 06 May 2026 11:08:41 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:33 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/PbaraXgyBf5cecgbD4TJLU-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[An illustration showing an AI agent side profile, depicted as a blue robot, with seven human faces in varying earthy metallic tones shown to the right.]]></media:description>                                                            <media:text><![CDATA[An illustration showing an AI agent side profile, depicted as a blue robot, with seven human faces in varying earthy metallic tones shown to the right.]]></media:text>
                                <media:title type="plain"><![CDATA[An illustration showing an AI agent side profile, depicted as a blue robot, with seven human faces in varying earthy metallic tones shown to the right.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/PbaraXgyBf5cecgbD4TJLU-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>UK consumers are growing increasingly concerned over AI-related controls, accountability, and governance, according to new research from EY. </p><p>The consultancy’s 2026 AI Sentiment Index found that nearly three-quarters (74%) of Brits have used AI recently, with the technology now deeply integrated into daily life and work. </p><p>Only 14% are comfortable with fully autonomous AI solutions, however, highlighting a growing demand for oversight. Just 43% of respondents said they trust companies using AI-powered services, while only 41% trust government agencies to manage AI-related data effectively. </p><p>It’s those who use AI most frequently, particularly educated, white‑collar workers, that are most concerned.</p><p>“AI adoption in the UK is rapidly advancing, but trust is not keeping pace with technological capability. Whilst consumers are engaging with AI every day, many still want greater clarity about who is accountable when decisions are made on their behalf," said Matthew Ringelheim, EY UK and Ireland AI leader.</p><p>“This is a critical moment for organizations. As AI systems become more autonomous, trust must be embedded through strong data foundations, clear accountability and visible human oversight. Our research shows UK users want greater control and transparency, reinforcing the need to move beyond AI adoption for its own sake."</p><h2 id="ai-benefits-are-becoming-clear">AI benefits are becoming clear</h2><p>While 73% of respondents are concerned about AI systems being hacked or breached, they're happy to engage with AI when they see clear benefits. </p><p>This tends to mean using AI in practical, outcome‑driven contexts, particularly where systems can improve response times (59%) and reliability (35%), or create better value for money (52%).</p><p>In more emotionally sensitive or high stakes situations, EY found users seem more comfortable using AI where it’s clear that safeguards, privacy, and consistent decision-making are in place. </p><p>Half of all UK respondents, for example, said they had consciously used health or wellness applications in the last six months, while 35% used it for financial activities where consistency and data protection are critical.</p><p>The message for businesses is that encouraging more use of AI will depend on reassuring users that strong governance measures are in place. </p><h2 id="skills-and-awareness-are-key">Skills and awareness are key</h2><p>Elsewhere in the survey, EY noted that UK users reported a lack of appropriate skills and training in the technology. Just 23% of respondents said they had received significant <a href="https://www.itpro.com/technology/artificial-intelligence/aws-to-give-ai-skills-to-100-000-people-in-the-uk-by-2030">training or education on AI</a> in the workplace. </p><p>Ringelheim said skills development plays a “critical role” in both bolstering adoption rates, but also building trust in the systems used on a daily basis. </p><p>“As <a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> become more capable, people will need greater confidence in how they’re used at work and clearer, practical guidance on how to use them responsibly," he commented. </p><p>"Training also better equips users to better spot errors, challenge outputs and make more informed decisions on when to rely on AI and when to escalate human judgement. Workforce confidence — built through the right skills — will be decisive in turning AI momentum into long‑term growth for the UK.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Five Eyes agencies sound alarm over risky agentic AI deployments ]]></title>
                                                                                                <dc:content><![CDATA[ <p><a href="https://www.itpro.com/technology/artificial-intelligence/practical-ai-the-age-of-agentic-ai">Agentic AI</a> is so risky it may be worth finding other ways to automate tasks  and only use the technology for non-sensitive tasks. </p><p>That's according to the Five Eyes group of intelligence agencies, made up of <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>centers from Australia, the US, Canada, the UK, and New Zealand. </p><p>The agencies have published a <a href="https://www.cyber.gov.au/business-government/secure-design/artificial-intelligence/careful-adoption-of-agentic-ai-services" target="_blank"><u>report</u></a> warning about the risks of using agentic AI in businesses, and the short version is only use it where strictly necessary and do so with caution. Indeed, they even pointed out that there's safer ways to manage work. </p><p>"Where possible, organizations should also consider a full spectrum of solutions for repetitive tasks, including reducing or eliminating low-value processes, which may be lower risk compared to agentic AI solutions," the report said. </p><p> "Until security practices, evaluation methods and standards mature, organisations should assume that agentic AI systems may behave unexpectedly and plan deployments accordingly, prioritizing resilience, reversibility and risk containment over efficiency gains,” the report added.</p><p>Industry experts have long been <a href="https://www.itpro.com/security/enterprises-are-adopting-agents-faster-than-they-can-secure-and-govern-them-experts-warn-its-a-disaster-waiting-to-happen"><u>concerned about the security and operational risks</u></a> of using AI, and earlier this month PocketOS offered a timely example when an <a href="https://www.itpro.com/technology/artificial-intelligence/nine-seconds-was-all-it-took-for-an-ai-agent-to-wipe-a-startups-database-experts-warn-its-a-glimpse-into-the-future-challenges-of-identity-security"><u>AI agent wiped out the startup's production database</u></a> in just nine seconds. </p><h2 id="is-agentic-ai-risky">Is agentic AI risky?</h2><p>Generally, introducing agentic AI comes with several concerns, including the potential for increased attack surfaces and greater complexity, as well as data protection risks. </p><p>But those challenges are difficult to address as models may evolve during deployment, or simply take a different route to a result than previously tested, while the inability to see what part of a system made a decision raises accountability issues. </p><p>"Gaps in agentic AI cyber security tooling and the immaturity of relevant standards further amplify these risks," the report added. "Governance mechanisms designed for human actors do not always translate effectively to autonomous AI agents."</p><iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/" src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/"></iframe><p>More specifically, the report noted that agentic AI raises risks around privileges. For example, by giving an agent broad access to do its job, any malicious actor that takes it over inherits its "excessive privileges". </p><p>AI's erratic behaviour is another risk, perhaps meeting requests in unexpected ways or finding technical loopholes. One example is specification gaming, the report noted. </p><p>"For example, an AI agent tasked with maximizing system uptime might disable security updates to avoid reboots."</p><p>The report also warned that agents can over-optimize and push boundaries, misinterpret requests, and be outright deceptive. </p><p>"As AI systems become more sophisticated, they may develop capabilities that designers did not explicitly program or anticipate," the report added. </p><p>"Complex AI models interacting with real-world systems can display behaviours that even their creators did not foresee. This unpredictability makes it difficult to assess security risks fully before deployment."</p><p>Beyond that, <a href="https://www.itpro.com/security/cyber-crime/agentic-ai-cybersecurity-risks">agents are a key target for hackers</a> because of their access to data as well as corporate systems. </p><h2 id="security-best-practices-are-critical">Security best practices are critical</h2><p>For those businesses using agentic AI, precautions are necessary, the report warned. </p><p>"Agentic AI developers, vendors and operators should implement a layered defence and strict access controls to reduce the likelihood of compromise," the report noted. </p><p>Defense should, as ever, begin at the design stage, the report noted – not crammed in just before implementation.</p><p>"Careful consideration of the system architecture, including security controls and tooling, is necessary," the report said. </p><p>"Practitioners should understand threats, anticipate risks to agentic AI systems and proactively integrate mitigations into system design before development and deployment."</p><p>The Five Eyes agencies in particular warned about controlling prompt contexts, to limit the damage of prompt injections, and ensure oversight mechanisms are built in with a human in the loop to avoid issues and boost trust. </p><p>Strong <a href="https://www.itpro.com/strategy/28935/what-is-identity-management-and-what-role-does-it-play-in-security-strategy">identity management</a> is also necessary, with the report recommending developers construct each agent as its own "distinct principle" with its own unique keys and certificates, blocking access to any unknown agent or key. </p><p>Elsewhere, the report reminded businesses of the need for defense in depth, noting that any single precaution may fail, whether the attack vector is AI or otherwise. </p><p>Beyond those recommendations, the report called for AI developers to harden agents against attacks or weakness with comprehensive testing and evaluation, using adversarial testing and simulated environments, with an eye to future risks. </p><p>"AI agents operate autonomously in complex environments and therefore require more thorough evaluations than LLMs," the report said. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/five-eyes-agencies-sound-alarm-over-risky-agentic-ai-deployments</link>
                                                                            <description>
                            <![CDATA[ Security agencies have urged organizations to establish clear boundaries and guardrails for AI agents ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">MiufYjDJwbR3ZhKEHZpEL7</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/eA94AYk6DLNFxKRQtGuKoT-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 05 May 2026 10:44:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:41 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/eA94AYk6DLNFxKRQtGuKoT-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[AI concept image showing digitized human eye monitoring digital interfaces and software.]]></media:description>                                                            <media:text><![CDATA[AI concept image showing digitized human eye monitoring digital interfaces and software.]]></media:text>
                                <media:title type="plain"><![CDATA[AI concept image showing digitized human eye monitoring digital interfaces and software.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/eA94AYk6DLNFxKRQtGuKoT-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p><a href="https://www.itpro.com/technology/artificial-intelligence/practical-ai-the-age-of-agentic-ai">Agentic AI</a> is so risky it may be worth finding other ways to automate tasks  and only use the technology for non-sensitive tasks. </p><p>That's according to the Five Eyes group of intelligence agencies, made up of <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>centers from Australia, the US, Canada, the UK, and New Zealand. </p><p>The agencies have published a <a href="https://www.cyber.gov.au/business-government/secure-design/artificial-intelligence/careful-adoption-of-agentic-ai-services" target="_blank"><u>report</u></a> warning about the risks of using agentic AI in businesses, and the short version is only use it where strictly necessary and do so with caution. Indeed, they even pointed out that there's safer ways to manage work. </p><p>"Where possible, organizations should also consider a full spectrum of solutions for repetitive tasks, including reducing or eliminating low-value processes, which may be lower risk compared to agentic AI solutions," the report said. </p><p> "Until security practices, evaluation methods and standards mature, organisations should assume that agentic AI systems may behave unexpectedly and plan deployments accordingly, prioritizing resilience, reversibility and risk containment over efficiency gains,” the report added.</p><p>Industry experts have long been <a href="https://www.itpro.com/security/enterprises-are-adopting-agents-faster-than-they-can-secure-and-govern-them-experts-warn-its-a-disaster-waiting-to-happen"><u>concerned about the security and operational risks</u></a> of using AI, and earlier this month PocketOS offered a timely example when an <a href="https://www.itpro.com/technology/artificial-intelligence/nine-seconds-was-all-it-took-for-an-ai-agent-to-wipe-a-startups-database-experts-warn-its-a-glimpse-into-the-future-challenges-of-identity-security"><u>AI agent wiped out the startup's production database</u></a> in just nine seconds. </p><h2 id="is-agentic-ai-risky">Is agentic AI risky?</h2><p>Generally, introducing agentic AI comes with several concerns, including the potential for increased attack surfaces and greater complexity, as well as data protection risks. </p><p>But those challenges are difficult to address as models may evolve during deployment, or simply take a different route to a result than previously tested, while the inability to see what part of a system made a decision raises accountability issues. </p><p>"Gaps in agentic AI cyber security tooling and the immaturity of relevant standards further amplify these risks," the report added. "Governance mechanisms designed for human actors do not always translate effectively to autonomous AI agents."</p><iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/" src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/"></iframe><p>More specifically, the report noted that agentic AI raises risks around privileges. For example, by giving an agent broad access to do its job, any malicious actor that takes it over inherits its "excessive privileges". </p><p>AI's erratic behaviour is another risk, perhaps meeting requests in unexpected ways or finding technical loopholes. One example is specification gaming, the report noted. </p><p>"For example, an AI agent tasked with maximizing system uptime might disable security updates to avoid reboots."</p><p>The report also warned that agents can over-optimize and push boundaries, misinterpret requests, and be outright deceptive. </p><p>"As AI systems become more sophisticated, they may develop capabilities that designers did not explicitly program or anticipate," the report added. </p><p>"Complex AI models interacting with real-world systems can display behaviours that even their creators did not foresee. This unpredictability makes it difficult to assess security risks fully before deployment."</p><p>Beyond that, <a href="https://www.itpro.com/security/cyber-crime/agentic-ai-cybersecurity-risks">agents are a key target for hackers</a> because of their access to data as well as corporate systems. </p><h2 id="security-best-practices-are-critical">Security best practices are critical</h2><p>For those businesses using agentic AI, precautions are necessary, the report warned. </p><p>"Agentic AI developers, vendors and operators should implement a layered defence and strict access controls to reduce the likelihood of compromise," the report noted. </p><p>Defense should, as ever, begin at the design stage, the report noted – not crammed in just before implementation.</p><p>"Careful consideration of the system architecture, including security controls and tooling, is necessary," the report said. </p><p>"Practitioners should understand threats, anticipate risks to agentic AI systems and proactively integrate mitigations into system design before development and deployment."</p><p>The Five Eyes agencies in particular warned about controlling prompt contexts, to limit the damage of prompt injections, and ensure oversight mechanisms are built in with a human in the loop to avoid issues and boost trust. </p><p>Strong <a href="https://www.itpro.com/strategy/28935/what-is-identity-management-and-what-role-does-it-play-in-security-strategy">identity management</a> is also necessary, with the report recommending developers construct each agent as its own "distinct principle" with its own unique keys and certificates, blocking access to any unknown agent or key. </p><p>Elsewhere, the report reminded businesses of the need for defense in depth, noting that any single precaution may fail, whether the attack vector is AI or otherwise. </p><p>Beyond those recommendations, the report called for AI developers to harden agents against attacks or weakness with comprehensive testing and evaluation, using adversarial testing and simulated environments, with an eye to future risks. </p><p>"AI agents operate autonomously in complex environments and therefore require more thorough evaluations than LLMs," the report said. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ UK firms accelerate ‘sovereign AI’ plans amid concerns over dependence on overseas tech ]]></title>
                                                                                                <dc:content><![CDATA[ <p>More than two-thirds of UK companies have an exit strategy prepared if their primary AI provider restricts access – a concern that's increasingly driving companies to prioritize <a href="https://www.itpro.com/security/data-protection/lets-talk-about-digital-sovereignty">digital sovereignty</a>.</p><p>That's according to a survey of 500 IT decision makers across the UK, Netherlands, France, Germany, and Italy by Red Hat, which found plans for digital or <a href="https://www.itpro.com/infrastructure/ai-infrastructure-global-divide">AI sovereignty</a> are no longer aspirational but an operational priority.</p><p>While 67% of UK companies are prepping for the worst case scenario, the Red Hat study suggested that 43% of those organizations would still face moderate to significant business impact if a primary AI provider restricted service. </p><p>“AI platforms are increasingly part of UK organizations’ critical infrastructure," said Joanna Hodgson, Country Manager for the UK at Red Hat. "Many have written exit strategies in preparation for any challenges, but our survey shows that actually executing a switch without disruption remains difficult."</p><p>Red Hat noted that 87% of UK companies are already using agentic AI systems – slightly behind France at 91% and Germany at 90% – but while they're using AI, they don't have key business continuity considerations in place yet. </p><p>Just 25% reported having strong governance frameworks in place for how they used agentic AI, with 17% saying such governance is "basic or minimal". Beyond that, fewer than half (48%) of UK companies said they had complete visibility of where their data is stored, processed, and potentially accessible, suggesting full AI sovereignty is some way off. </p><p>"To close that gap, enterprises need greater control over how and where AI runs, and a consistent way to govern fast‑moving technologies like agentic AI," she added. </p><p>The Red Hat survey marks the latest in a string of studies highlighting increased efforts to shore up digital sovereignty across Europe. <a href="https://www.itpro.com/infrastructure/sovereign-infrastructure-spend-to-triple-in-europe-as-fifth-of-workloads-stay-local"><u>Research from Gartner</u></a>, for example, predicted sovereign cloud spend will triple in the next few years. </p><p>This increased investment is, at least in part, being driven by geopolitical developments that have sparked concerns about an overreliance on American tech giants. </p><p>Last week, the <a href="https://www.itpro.com/infrastructure/uk-too-reliant-on-big-tech-needs-prioritise-digital-sovereignty-say-rights-organizers"><u>Open Rights Group called for a "strategic shift"</u></a> in the UK to help boost spending on local tech firms while avoiding vendor lock-in and security or surveillance risks. </p><h2 id="is-open-source-ai-the-answer">Is open source AI the answer?</h2><p>The Red Hat survey found strong support for government-driven AI sovereignty, in particular focusing on open source principles. Nine-in-ten respondents said that public policy and regulation should "actively mandate" open source principles as a route to achieving AI sovereignty. </p><p>"Enterprise open source gives UK businesses the transparency, flexibility and shared innovation they need to treat AI as a resilient, sovereign capability," Hodgson said. </p><p>A <a href="https://www.itpro.com/technology/artificial-intelligence/eu-businesses-will-flock-to-region-specific-ai-platforms-by-2027-but-cost-could-be-a-major-hurdle"><u>recent study from Gartner</u></a> found that roughly one-third of countries will introduce region-specific AI platforms within two years in a bid to shore up data sovereignty efforts. </p><p>That process could result in higher costs, however, with nations opting for this route forced to spend at least 1% of GDP to expand AI infrastructure capabilities. </p><h2 id="freedom-to-choose">Freedom to choose</h2><p>The survey results suggest companies are looking for more flexibility in their AI stack and want to keep control, according to Hans Roth, Red Hat’s senior vice president and general manager for EMEA.</p><p>"Across EMEA, boardroom conversation has moved beyond experimentation to how AI can be deployed in a way that meets sovereignty, security and regulatory expectations," Roth said. </p><p>"The survey results show strong <a href="https://www.itpro.com/software/open-source/the-pros-and-cons-of-open-source-ai-for-business"><u>support for open source principles</u></a> and for clear policy frameworks that embed transparency and auditability into AI." </p><p>Roth added that organisations aren’t looking for “another closed, one-size-fits-all stack” and value the flexibility to combine different models and cloud services while retaining strict control. </p><p>However, companies need to be <a href="https://www.itpro.com/software/open-source/the-pros-and-cons-of-open-source-ai-for-business"><u>wary of "open washing",</u></a> a common issue in open source AI, where openness can often refer only to specific weights, rather than more widely. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/uk-firms-accelerate-sovereign-ai-plans-amid-concerns-over-dependence-on-overseas-tech</link>
                                                                            <description>
                            <![CDATA[ A Red Hat report shows firms are prioritizing sovereign AI over fears that foreign providers could restrict access ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">BENUWE2qrda4t9r9iDDm6J</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/XUFTVxXHrTRAVBdujTYj6P-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 24 Apr 2026 10:31:30 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:41 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Nicole Kobie ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/8Y8JDDTQ7XDEk49FoAFP2S-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Nicole Kobie first started writing for ITPro in 2007. As a freelance journalist covering technology and business, Nicole&#039;s work includes  bylines in New Scientist, Wired, PC Pro and many more. &lt;/p&gt;&lt;p&gt;Nicole the author of a book about the history of technology, The Long History of the Future.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/XUFTVxXHrTRAVBdujTYj6P-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Artificial intelligence (AI) concept image showing a digitized cube with &#039;AI&#039; written on it resting on top of circuit boards.]]></media:description>                                                            <media:text><![CDATA[Artificial intelligence (AI) concept image showing a digitized cube with &#039;AI&#039; written on it resting on top of circuit boards.]]></media:text>
                                <media:title type="plain"><![CDATA[Artificial intelligence (AI) concept image showing a digitized cube with &#039;AI&#039; written on it resting on top of circuit boards.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/XUFTVxXHrTRAVBdujTYj6P-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>More than two-thirds of UK companies have an exit strategy prepared if their primary AI provider restricts access – a concern that's increasingly driving companies to prioritize <a href="https://www.itpro.com/security/data-protection/lets-talk-about-digital-sovereignty">digital sovereignty</a>.</p><p>That's according to a survey of 500 IT decision makers across the UK, Netherlands, France, Germany, and Italy by Red Hat, which found plans for digital or <a href="https://www.itpro.com/infrastructure/ai-infrastructure-global-divide">AI sovereignty</a> are no longer aspirational but an operational priority.</p><p>While 67% of UK companies are prepping for the worst case scenario, the Red Hat study suggested that 43% of those organizations would still face moderate to significant business impact if a primary AI provider restricted service. </p><p>“AI platforms are increasingly part of UK organizations’ critical infrastructure," said Joanna Hodgson, Country Manager for the UK at Red Hat. "Many have written exit strategies in preparation for any challenges, but our survey shows that actually executing a switch without disruption remains difficult."</p><p>Red Hat noted that 87% of UK companies are already using agentic AI systems – slightly behind France at 91% and Germany at 90% – but while they're using AI, they don't have key business continuity considerations in place yet. </p><p>Just 25% reported having strong governance frameworks in place for how they used agentic AI, with 17% saying such governance is "basic or minimal". Beyond that, fewer than half (48%) of UK companies said they had complete visibility of where their data is stored, processed, and potentially accessible, suggesting full AI sovereignty is some way off. </p><p>"To close that gap, enterprises need greater control over how and where AI runs, and a consistent way to govern fast‑moving technologies like agentic AI," she added. </p><p>The Red Hat survey marks the latest in a string of studies highlighting increased efforts to shore up digital sovereignty across Europe. <a href="https://www.itpro.com/infrastructure/sovereign-infrastructure-spend-to-triple-in-europe-as-fifth-of-workloads-stay-local"><u>Research from Gartner</u></a>, for example, predicted sovereign cloud spend will triple in the next few years. </p><p>This increased investment is, at least in part, being driven by geopolitical developments that have sparked concerns about an overreliance on American tech giants. </p><p>Last week, the <a href="https://www.itpro.com/infrastructure/uk-too-reliant-on-big-tech-needs-prioritise-digital-sovereignty-say-rights-organizers"><u>Open Rights Group called for a "strategic shift"</u></a> in the UK to help boost spending on local tech firms while avoiding vendor lock-in and security or surveillance risks. </p><h2 id="is-open-source-ai-the-answer">Is open source AI the answer?</h2><p>The Red Hat survey found strong support for government-driven AI sovereignty, in particular focusing on open source principles. Nine-in-ten respondents said that public policy and regulation should "actively mandate" open source principles as a route to achieving AI sovereignty. </p><p>"Enterprise open source gives UK businesses the transparency, flexibility and shared innovation they need to treat AI as a resilient, sovereign capability," Hodgson said. </p><p>A <a href="https://www.itpro.com/technology/artificial-intelligence/eu-businesses-will-flock-to-region-specific-ai-platforms-by-2027-but-cost-could-be-a-major-hurdle"><u>recent study from Gartner</u></a> found that roughly one-third of countries will introduce region-specific AI platforms within two years in a bid to shore up data sovereignty efforts. </p><p>That process could result in higher costs, however, with nations opting for this route forced to spend at least 1% of GDP to expand AI infrastructure capabilities. </p><h2 id="freedom-to-choose">Freedom to choose</h2><p>The survey results suggest companies are looking for more flexibility in their AI stack and want to keep control, according to Hans Roth, Red Hat’s senior vice president and general manager for EMEA.</p><p>"Across EMEA, boardroom conversation has moved beyond experimentation to how AI can be deployed in a way that meets sovereignty, security and regulatory expectations," Roth said. </p><p>"The survey results show strong <a href="https://www.itpro.com/software/open-source/the-pros-and-cons-of-open-source-ai-for-business"><u>support for open source principles</u></a> and for clear policy frameworks that embed transparency and auditability into AI." </p><p>Roth added that organisations aren’t looking for “another closed, one-size-fits-all stack” and value the flexibility to combine different models and cloud services while retaining strict control. </p><p>However, companies need to be <a href="https://www.itpro.com/software/open-source/the-pros-and-cons-of-open-source-ai-for-business"><u>wary of "open washing",</u></a> a common issue in open source AI, where openness can often refer only to specific weights, rather than more widely. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Anthropic toned down Opus 4.7’s cyber capabilities in wake of Claude Mythos release ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Anthropic claims it used new techniques to “differentially reduce” the cyber capabilities of its Opus 4.7 model in the wake of the <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">Claude Mythos release</a>. </p><p>The company unveiled the new model on 16 April, providing users with marked improvements in areas such as <a href="https://www.itpro.com/technology/artificial-intelligence/ai-in-software-engineering-six-ways-the-profession-is-changing">software engineering</a>. The AI developer specifically highlighted gains in coding, as well as improvements on knowledge work tasks. </p><p>“Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks,” the company said in a <a href="https://www.anthropic.com/news/claude-opus-4-7" target="_blank"><u>blog post</u></a>. </p><p>“Users report being able to hand off their hardest coding work—the kind that previously needed close supervision—to Opus 4.7 with confidence.”</p><p>Anthropic noted that Opus 4.7 is capable of handling “complex, long-running tasks with rigor and consistency” and is “substantially better” at following instructions. </p><h2 id="reduced-cyber-capabilities">Reduced cyber capabilities</h2><p>The launch of Opus 4.7 comes hot on the heels of the gated release of Claude Mythos, the company’s most powerful model yet which it claims “reshape <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity</a>”. </p><p>Through <a href="https://www.itpro.com/security/ai-is-raising-the-stakes-for-cyber-professionals-claude-mythos-just-took-things-to-another-level">Project Glasswing</a>, a host of big tech partners have been given preview access to the model in a bid to test and evaluate its defensive capabilities, which initial testing shows could deliver marked improvements. </p><p>Anthropic claimed Mythos identified “thousands of zero-day vulnerabilities” during testing, many of which were critical and some that had flown under the radar for nearly three decades. </p><p>Concerns have been raised over its potential for nefarious use, however, and Anthropic said the limited release aimed to prevent misuse by threat actors. </p><p>Opus 4.7 is the first model released by the firm in the wake of Project Glasswing, and Anthropic noted that its cyber capabilities are “not as advanced” as those of Mythos. </p><p>Indeed, it appears the company actively toned down its capabilities in cybersecurity use-cases. </p><p>“During its training we experimented with efforts to differentially reduce these capabilities,” the company said in a blog post. </p><p>Opus 4.7 also comes with new safeguards that detect and block requests that “indicate prohibited or high-risk cybersecurity uses”. Long-term, the testing of these safeguards could help steer the company’s approach to a full public release of Mythos. </p><p>“What we learn from the real-world deployment of these safeguards will help us work towards our eventual goal of a broad release of Mythos-class models,” the company said. </p><p>“Security professionals who wish to use Opus 4.7 for legitimate cybersecurity purposes (such as vulnerability research, penetration testing, and red-teaming) are invited to join our new Cyber Verification Program.”</p><h2 id="will-anthropic-s-gambit-work">Will Anthropic's gambit work?</h2><p>Anthropic is working at speed, with just over two months having passed between the release of its last Opus model and this successor. The headline here is, of course, the leap in code generation and agentic computer use – an unsurprising focus for a company that has become widely known for <a href="https://www.itpro.com/security/data-breaches/there-was-a-manual-deploy-step-that-should-have-been-better-automated-claude-code-creator-confirms-cause-of-massive-source-code-leak"><u>Claude Code</u></a> and <a href="https://www.itpro.com/technology/artificial-intelligence/everything-you-need-to-know-about-anthropic-claude-cowork"><u>Claude Cowork</u></a>.</p><p>At the same time, Anthropic appears to be more and more concerned about releasing cyber-capable models to the general public. I’m interested in seeing actual testing of Opus 4.7’s cyber safeguards, as hiding capabilities behind model refusal is never a guaranteed method to stop bad actors from accessing them via <a href="https://www.itpro.com/security/ncsc-issues-urgent-warning-over-growing-ai-prompt-injection-risks-heres-what-you-need-to-know"><u>prompt injection</u></a>.</p><p>Interestingly, Anthropic says it’s paired these traditional guardrails with active efforts to reduce Opus 4.7’s cybersecurity performance during training. This is the first time I can remember a model developer actively trying to scuttle one of its frontier models.</p><p>If it proves successful, the ‘Opus’ family could become more fragmented, with sub-models that specialize in certain tasks and fall behind on others. </p><p>For example, as it leans further into computer use capabilities with Claude Cowork, it could be beneficial for the firm to produce a model that excels at this but not code generation, to remove the possibility of malicious misuse.</p><p>For now, it seems, Anthropic is saving its best cyber performance for Mythos. That said, we have to be careful about taking everything Anthropic says about Mythos at face value. </p><p>Recent analysis from the UK’s <a href="https://www.itpro.com/security/there-needs-to-be-an-order-of-magnitude-more-effort-ai-security-experts-call-for-focused-evaluation-of-frontier-models-and-agentic-systems"><u>AI Security Institute</u></a>, which rigorously tested Claude Mythos’ capabilities using custom cybersecurity tests, found that Mythos Preview is more capable than previous frontier models in cyber performance but doesn’t exceed human cyber performance.</p><p>In a 32-step corporate network attack simulation, Mythos Preview was the first model to reach the end, albeit only in three of its ten attempts.</p><p>“Mythos Preview’s success on one cyber range indicates that it is at least capable of autonomously attacking small, weakly defended and vulnerable enterprise systems where access to a network has been gained,” AISI researchers wrote.</p><p>“However, our ranges have important differences from real-world environments that make them easier targets,” they added, explaining this includes a lack of active defenders, defensive tooling, and security alerts that would be present in the real world.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/anthropic-claude-opus-claude-mythos-cyber-capabilities</link>
                                                                            <description>
                            <![CDATA[ Testing of new cyber-related safeguards for Anthropic’s Opus 4.7 model could shape the future public release of Claude Mythos ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">qb2kvfvZbt46Hq3C46oKsT</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/ixBAEHkHMhpfNLKPfvD29-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 17 Apr 2026 09:52:43 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:13 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                        <dc:contributor><![CDATA[ Rory Bathgate ]]></dc:contributor>
                                                                    <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/ixBAEHkHMhpfNLKPfvD29-1920-80.jpg">
                                                            <media:credit><![CDATA[Anthropic]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Claude Opus 4.7 logo and branding featuring a human head on left hand side and notepad on right.]]></media:description>                                                            <media:text><![CDATA[Claude Opus 4.7 logo and branding featuring a human head on left hand side and notepad on right.]]></media:text>
                                <media:title type="plain"><![CDATA[Claude Opus 4.7 logo and branding featuring a human head on left hand side and notepad on right.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/ixBAEHkHMhpfNLKPfvD29-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Anthropic claims it used new techniques to “differentially reduce” the cyber capabilities of its Opus 4.7 model in the wake of the <a href="https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity">Claude Mythos release</a>. </p><p>The company unveiled the new model on 16 April, providing users with marked improvements in areas such as <a href="https://www.itpro.com/technology/artificial-intelligence/ai-in-software-engineering-six-ways-the-profession-is-changing">software engineering</a>. The AI developer specifically highlighted gains in coding, as well as improvements on knowledge work tasks. </p><p>“Opus 4.7 is a notable improvement on Opus 4.6 in advanced software engineering, with particular gains on the most difficult tasks,” the company said in a <a href="https://www.anthropic.com/news/claude-opus-4-7" target="_blank"><u>blog post</u></a>. </p><p>“Users report being able to hand off their hardest coding work—the kind that previously needed close supervision—to Opus 4.7 with confidence.”</p><p>Anthropic noted that Opus 4.7 is capable of handling “complex, long-running tasks with rigor and consistency” and is “substantially better” at following instructions. </p><h2 id="reduced-cyber-capabilities">Reduced cyber capabilities</h2><p>The launch of Opus 4.7 comes hot on the heels of the gated release of Claude Mythos, the company’s most powerful model yet which it claims “reshape <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity</a>”. </p><p>Through <a href="https://www.itpro.com/security/ai-is-raising-the-stakes-for-cyber-professionals-claude-mythos-just-took-things-to-another-level">Project Glasswing</a>, a host of big tech partners have been given preview access to the model in a bid to test and evaluate its defensive capabilities, which initial testing shows could deliver marked improvements. </p><p>Anthropic claimed Mythos identified “thousands of zero-day vulnerabilities” during testing, many of which were critical and some that had flown under the radar for nearly three decades. </p><p>Concerns have been raised over its potential for nefarious use, however, and Anthropic said the limited release aimed to prevent misuse by threat actors. </p><p>Opus 4.7 is the first model released by the firm in the wake of Project Glasswing, and Anthropic noted that its cyber capabilities are “not as advanced” as those of Mythos. </p><p>Indeed, it appears the company actively toned down its capabilities in cybersecurity use-cases. </p><p>“During its training we experimented with efforts to differentially reduce these capabilities,” the company said in a blog post. </p><p>Opus 4.7 also comes with new safeguards that detect and block requests that “indicate prohibited or high-risk cybersecurity uses”. Long-term, the testing of these safeguards could help steer the company’s approach to a full public release of Mythos. </p><p>“What we learn from the real-world deployment of these safeguards will help us work towards our eventual goal of a broad release of Mythos-class models,” the company said. </p><p>“Security professionals who wish to use Opus 4.7 for legitimate cybersecurity purposes (such as vulnerability research, penetration testing, and red-teaming) are invited to join our new Cyber Verification Program.”</p><h2 id="will-anthropic-s-gambit-work">Will Anthropic's gambit work?</h2><p>Anthropic is working at speed, with just over two months having passed between the release of its last Opus model and this successor. The headline here is, of course, the leap in code generation and agentic computer use – an unsurprising focus for a company that has become widely known for <a href="https://www.itpro.com/security/data-breaches/there-was-a-manual-deploy-step-that-should-have-been-better-automated-claude-code-creator-confirms-cause-of-massive-source-code-leak"><u>Claude Code</u></a> and <a href="https://www.itpro.com/technology/artificial-intelligence/everything-you-need-to-know-about-anthropic-claude-cowork"><u>Claude Cowork</u></a>.</p><p>At the same time, Anthropic appears to be more and more concerned about releasing cyber-capable models to the general public. I’m interested in seeing actual testing of Opus 4.7’s cyber safeguards, as hiding capabilities behind model refusal is never a guaranteed method to stop bad actors from accessing them via <a href="https://www.itpro.com/security/ncsc-issues-urgent-warning-over-growing-ai-prompt-injection-risks-heres-what-you-need-to-know"><u>prompt injection</u></a>.</p><p>Interestingly, Anthropic says it’s paired these traditional guardrails with active efforts to reduce Opus 4.7’s cybersecurity performance during training. This is the first time I can remember a model developer actively trying to scuttle one of its frontier models.</p><p>If it proves successful, the ‘Opus’ family could become more fragmented, with sub-models that specialize in certain tasks and fall behind on others. </p><p>For example, as it leans further into computer use capabilities with Claude Cowork, it could be beneficial for the firm to produce a model that excels at this but not code generation, to remove the possibility of malicious misuse.</p><p>For now, it seems, Anthropic is saving its best cyber performance for Mythos. That said, we have to be careful about taking everything Anthropic says about Mythos at face value. </p><p>Recent analysis from the UK’s <a href="https://www.itpro.com/security/there-needs-to-be-an-order-of-magnitude-more-effort-ai-security-experts-call-for-focused-evaluation-of-frontier-models-and-agentic-systems"><u>AI Security Institute</u></a>, which rigorously tested Claude Mythos’ capabilities using custom cybersecurity tests, found that Mythos Preview is more capable than previous frontier models in cyber performance but doesn’t exceed human cyber performance.</p><p>In a 32-step corporate network attack simulation, Mythos Preview was the first model to reach the end, albeit only in three of its ten attempts.</p><p>“Mythos Preview’s success on one cyber range indicates that it is at least capable of autonomously attacking small, weakly defended and vulnerable enterprise systems where access to a network has been gained,” AISI researchers wrote.</p><p>“However, our ranges have important differences from real-world environments that make them easier targets,” they added, explaining this includes a lack of active defenders, defensive tooling, and security alerts that would be present in the real world.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Can Europe achieve AI sovereignty? ]]></title>
                                                                                                <dc:content><![CDATA[ <iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/fed30e50-bbea-4518-aab4-ed6d6bda41f7/" src="https://player.captivate.fm/episode/fed30e50-bbea-4518-aab4-ed6d6bda41f7/"></iframe><p>AI deployment is inextricably linked to the United States of America. Many of the major frontier labs are based there and the cloud providers that businesses depend on to access the latest models are majority American. </p><p>But all over Europe, businesses are considering questioning the extent to which their critical workloads should be dependent on the US.</p><p>In a recent edition of the podcast, we covered all things digital sovereignty. In this episode, we're zooming in on how this push for sovereignty is compatible with AI adoption for European businesses. What does ‘AI sovereignty’ mean in practical terms? And how can businesses isolate workloads from the wider world without falling behind?</p><p>In this episode, Rory speaks to René Büst, senior director, analyst in Gartner's Emerging Market Dynamics practice, to explore how European CIOs can balance sovereignty, agility, and compliance.</p><h2 id="highlights-3">Highlights</h2><p>"So if you look at the spectrum, we also see that the global providers are of course making some steps, because money talks in the end of the day, and they have opened here more sovereign offerings by, for example, partnering with local system integrators or local vendors."</p><p>"You can be 100% compliant, but at the same time 0% sovereign, because if someone else has a kill switch and can shut your environment off, you might be still compliant, but you won't exist anymore, right?"</p><p>"[T]here are more providers across Europe than I think the public knows about. And this is, of course, also an issue but this is absolutely one of the strategies that some have started to follow, or which which can be followed. But of course, it comes also with a cost, because one thing [that] is also true is that the more innovative AI models or AI technologies are still coming from the United States or from China. So there is a trade off, and CIOs then have to make compromise to say, 'Do I need really a lot of sovereignty, or can I balance the risks that way that I'm okay by using foreign AI technologies?'"</p><h2 id="footnotes">Footnotes</h2><ul><li><a href="https://www.itpro.com/infrastructure/can-the-uk-achieve-ai-sovereignty" target="_blank">Can the UK achieve AI sovereignty?</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/eu-businesses-will-flock-to-region-specific-ai-platforms-by-2027-but-cost-could-be-a-major-hurdle" target="_blank">EU businesses will flock to ‘region-specific’ AI platforms by 2027 – but cost could be a major hurdle</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/will-the-future-of-ai-be-made-in-europe-the-eu-thinks-so" target="_blank">Will the future of AI be made in Europe? The EU thinks so</a></li><li><a href="https://www.itpro.com/business/policy-and-legislation/the-second-enforcement-deadline-for-the-eu-ai-act-is-approaching-heres-what-businesses-need-to-know-about-the-general-purpose-ai-code-of-practice" target="_blank">The second enforcement deadline for the EU AI Act is approaching – here’s what businesses need to know about the General-Purpose AI Code of Practice</a></li><li><a href="https://www.itpro.com/business/policy-and-legislation/what-is-the-eus-ai-plan" target="_blank">What is the EU's AI plan?</a></li><li><a href="https://www.itpro.com/cloud/cloud-computing/aws-says-only-europeans-will-run-its-european-sovereign-cloud-service" target="_blank">AWS says only Europeans will run its European Sovereign Cloud service</a></li><li><a href="https://www.itpro.com/cloud/cloud-computing/google-is-getting-serious-on-cloud-sovereignty" target="_blank">Google is getting serious on cloud sovereignty</a></li><li><a href="https://www.itpro.com/cloud/cloud-computing/microsoft-sovereign-cloud-launch-eu-customers" target="_blank">What the new Microsoft Sovereign Cloud push means for European customers</a></li></ul><h2 id="subscribe">Subscribe </h2><ul><li><a href="https://apple.sjv.io/c/221109/473657/7613?subId1=itpro-gb-1243831151189624600&amp;sharedId=itpro-gb&amp;u=https%3A%2F%2Fpodcasts.apple.com%2Fgb%2Fpodcast%2Fthe-itpro-podcast%2Fid1483810154" target="_blank"><u>Subscribe to The IT Pro Podcast on Apple Podcasts</u></a></li><li><a href="https://open.spotify.com/show/7HpYehTy752KmtbwpOAgRZ" target="_blank"><u>Subscribe to The IT Pro Podcast on Spotify</u></a></li><li><a href="https://www.itpro.co.uk/newsletter-signup"><u>Subscribe to the IT Pro newsletter</u></a></li><li><a href="https://uk.linkedin.com/company/itpro-uk" target="_blank"><u>Join us on LinkedIn</u></a></li></ul> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/can-europe-achieve-ai-sovereignty</link>
                                                                            <description>
                            <![CDATA[ Efforts to run AI in trusted regions can clash with access to frontier model updates, business scalability ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">pbDPtXLFLSCCWWJsL6m6ad</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/qaVaBqV5c3P3xqFfR8Msi4-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 10 Apr 2026 06:30:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:22 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ rory.bathgate@futurenet.com (Rory Bathgate) ]]></author>                    <dc:creator><![CDATA[ Rory Bathgate ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Vep5JogbPhduK7R6CUWAm6-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Rory Bathgate was the Features and Multimedia Editor at ITPro until 2026, overseeing all in-depth content and case studies. Throughout his time at ITPro, Rory  charted the rise in popularity of generative AI and specifically companies such as Microsoft, OpenAI, and Google. &lt;/p&gt;&lt;p&gt;Alongside this, he delved into increasing calls for ethical and responsible AI as global legislators circle the technology, as well as the latest in mobile networking technology, from 5G mmWave to the 3G sunset and how it will affect businesses.&lt;/p&gt;&lt;p&gt;He  provided coverage from high-profile tech conferences such as Dell Technologies World, SuiteWorld, and VMware Explore Europe. His on-the-ground coverage  included live blogs, extensive daily coverage of the most significant announcements, analysis pieces, and podcast.&lt;/p&gt;&lt;p&gt;Rory was the co-host of the ITPro Podcast alongside Jane McCallion, where he swapped a keyboard for a microphone to discuss the latest learnings in tech.&lt;/p&gt;&lt;p&gt;Rory has an MA in Eighteenth-Century Studies from King’s College London, as well as a BA in English and American Literature from the University of Kent. He joined ITPro in 2022 as a graduate, after four years in student journalism.&lt;/p&gt;&lt;p&gt;In his free time, Rory enjoys photography and video editing, and can often be found at the cinema or reading a good science fiction paperback.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/qaVaBqV5c3P3xqFfR8Msi4-1920-80.jpg">
                                                            <media:credit><![CDATA[Future]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[The text &quot;Can Europe achieve AI sovereignty&quot; against a stylized AI chip on the EU flag with the the ITPro Podcast logo in the corner. The words &quot;AI sovereignty&quot; are in yellow, the rest are in white.]]></media:description>                                                            <media:text><![CDATA[The text &quot;Can Europe achieve AI sovereignty&quot; against a stylized AI chip on the EU flag with the the ITPro Podcast logo in the corner. The words &quot;AI sovereignty&quot; are in yellow, the rest are in white.]]></media:text>
                                <media:title type="plain"><![CDATA[The text &quot;Can Europe achieve AI sovereignty&quot; against a stylized AI chip on the EU flag with the the ITPro Podcast logo in the corner. The words &quot;AI sovereignty&quot; are in yellow, the rest are in white.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/qaVaBqV5c3P3xqFfR8Msi4-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/fed30e50-bbea-4518-aab4-ed6d6bda41f7/" src="https://player.captivate.fm/episode/fed30e50-bbea-4518-aab4-ed6d6bda41f7/"></iframe><p>AI deployment is inextricably linked to the United States of America. Many of the major frontier labs are based there and the cloud providers that businesses depend on to access the latest models are majority American. </p><p>But all over Europe, businesses are considering questioning the extent to which their critical workloads should be dependent on the US.</p><p>In a recent edition of the podcast, we covered all things digital sovereignty. In this episode, we're zooming in on how this push for sovereignty is compatible with AI adoption for European businesses. What does ‘AI sovereignty’ mean in practical terms? And how can businesses isolate workloads from the wider world without falling behind?</p><p>In this episode, Rory speaks to René Büst, senior director, analyst in Gartner's Emerging Market Dynamics practice, to explore how European CIOs can balance sovereignty, agility, and compliance.</p><h2 id="highlights-3">Highlights</h2><p>"So if you look at the spectrum, we also see that the global providers are of course making some steps, because money talks in the end of the day, and they have opened here more sovereign offerings by, for example, partnering with local system integrators or local vendors."</p><p>"You can be 100% compliant, but at the same time 0% sovereign, because if someone else has a kill switch and can shut your environment off, you might be still compliant, but you won't exist anymore, right?"</p><p>"[T]here are more providers across Europe than I think the public knows about. And this is, of course, also an issue but this is absolutely one of the strategies that some have started to follow, or which which can be followed. But of course, it comes also with a cost, because one thing [that] is also true is that the more innovative AI models or AI technologies are still coming from the United States or from China. So there is a trade off, and CIOs then have to make compromise to say, 'Do I need really a lot of sovereignty, or can I balance the risks that way that I'm okay by using foreign AI technologies?'"</p><h2 id="footnotes">Footnotes</h2><ul><li><a href="https://www.itpro.com/infrastructure/can-the-uk-achieve-ai-sovereignty" target="_blank">Can the UK achieve AI sovereignty?</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/eu-businesses-will-flock-to-region-specific-ai-platforms-by-2027-but-cost-could-be-a-major-hurdle" target="_blank">EU businesses will flock to ‘region-specific’ AI platforms by 2027 – but cost could be a major hurdle</a></li><li><a href="https://www.itpro.com/technology/artificial-intelligence/will-the-future-of-ai-be-made-in-europe-the-eu-thinks-so" target="_blank">Will the future of AI be made in Europe? The EU thinks so</a></li><li><a href="https://www.itpro.com/business/policy-and-legislation/the-second-enforcement-deadline-for-the-eu-ai-act-is-approaching-heres-what-businesses-need-to-know-about-the-general-purpose-ai-code-of-practice" target="_blank">The second enforcement deadline for the EU AI Act is approaching – here’s what businesses need to know about the General-Purpose AI Code of Practice</a></li><li><a href="https://www.itpro.com/business/policy-and-legislation/what-is-the-eus-ai-plan" target="_blank">What is the EU's AI plan?</a></li><li><a href="https://www.itpro.com/cloud/cloud-computing/aws-says-only-europeans-will-run-its-european-sovereign-cloud-service" target="_blank">AWS says only Europeans will run its European Sovereign Cloud service</a></li><li><a href="https://www.itpro.com/cloud/cloud-computing/google-is-getting-serious-on-cloud-sovereignty" target="_blank">Google is getting serious on cloud sovereignty</a></li><li><a href="https://www.itpro.com/cloud/cloud-computing/microsoft-sovereign-cloud-launch-eu-customers" target="_blank">What the new Microsoft Sovereign Cloud push means for European customers</a></li></ul><h2 id="subscribe">Subscribe </h2><ul><li><a href="https://apple.sjv.io/c/221109/473657/7613?subId1=itpro-gb-1243831151189624600&amp;sharedId=itpro-gb&amp;u=https%3A%2F%2Fpodcasts.apple.com%2Fgb%2Fpodcast%2Fthe-itpro-podcast%2Fid1483810154" target="_blank"><u>Subscribe to The IT Pro Podcast on Apple Podcasts</u></a></li><li><a href="https://open.spotify.com/show/7HpYehTy752KmtbwpOAgRZ" target="_blank"><u>Subscribe to The IT Pro Podcast on Spotify</u></a></li><li><a href="https://www.itpro.co.uk/newsletter-signup"><u>Subscribe to the IT Pro newsletter</u></a></li><li><a href="https://uk.linkedin.com/company/itpro-uk" target="_blank"><u>Join us on LinkedIn</u></a></li></ul>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Anthropic is worried hackers could abuse its Claude Mythos AI model ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Anthropic has announced the launch of a new scheme that will see the firm collaborate with an array of tech giants on the future of AI security. </p><p>Dubbed ‘Project Glasswing’, the initiative will bring together Amazon, Apple, Broadcom, Microsoft, Cisco, CrowdStrike, Palo Alto Networks, and the Linux Foundation to test a new AI model aimed at bolstering defensive cyber capabilities. </p><p>Anthropic said the model in question, Claude Mythos, has the potential to “reshape cybersecurity” and will only be available to organizations involved in the project. </p><p>As part of the scheme, launch partners will be given access to Mythos Preview for use in “defensive security work” and share learnings on the capabilities of the model. </p><p>Access has also been granted to 40 additional organisations that build or maintain critical software infrastructure in a bid to secure open source ecosystems, Anthropic revealed. </p><p>The company said it will commit up to $100 million in usage credits on this front, as well as a $4 million donation to open source security organizations. </p><h2 id="what-project-glasswing-aims-to-achieve">What Project Glasswing aims to achieve</h2><p>Project Glasswing stems from Anthropic’s own observations on the capabilities of Claude Mythos in recent weeks. Speculation over the launch of Mythos was fueled by a data leak which saw information on the general-purpose model exposed online.</p><p>That data leak pointed toward significant improvements in cybersecurity, which Anthropic has since confirmed. The company noted that it wasn’t trained specifically for <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>activities<a href="https://www.itpro.com/security/28133/what-is-cyber-security">, </a>but its capabilities are a “result of its strong agentic coding and reasoning skills”. </p><p>Simply put, Anthropic believes the model is powerful enough to warrant testing behind closed doors at trusted industry giants rather than releasing to the public, where it could be used for nefarious purposes. </p><p>“Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely,” the company said in a statement.</p><p>“The work of defending the world’s cyber infrastructure might take years; frontier AI capabilities are likely to advance substantially over just the next few months. For cyber defenders to come out ahead, we need to act now.”</p><h2 id="claude-mythos-testing-shows-solid-results">Claude Mythos testing shows solid results</h2><p>According to Anthropic, Mythos Preview has identified “thousands of zero-day vulnerabilities” in recent weeks, many of which were critical. </p><p>Notably, those identified by the model were “often subtle or difficult to detect” and dating back decades. The oldest flaw detected by Mythos included a 27-year old bug in the OpenBSD operating system. </p><p>“It also discovered a 16-year-old vulnerability in a widely used video software in a line of code that automated testing tools had hit five million times without ever catching the problem,” the company said. </p><p>Anthropic revealed it is also engaged in discussions with US government officials about Claude Mythos Preview and its “offensive and defensive cyber capabilities”.</p><h2 id="a-new-cybersecurity-paradigm">A new cybersecurity paradigm</h2><p>The launch of Project Glasswing is a seismic moment for Anthropic and one that could come to define the firm as it approaches a <a href="https://seekingalpha.com/article/4887778-anthropic-ipo-what-you-need-to-know" target="_blank"><u>potential IPO</u></a> later this year.</p><p>It’s worth underlining just how unusual it is for a tech company – let alone one that’s barely five years old – to announce a gated release of its software with such widespread endorsement. </p><p>While Google and Amazon are long-time backers of Anthropic, and Microsoft has recently joined in, the full-throated support of CrowdStrike and Palo Alto Networks reads to me as recognition that Claude Mythos Preview will usher in a serious leap in cybersecurity capabilities. </p><p>Here, cybersecurity companies that have long prided themselves on their proprietary <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI </a>are admitting that Anthropic’s latest release is already catching zero days that no other tools ever have.</p><p>We’ll have to wait and see just how effective it is and, with the release limited to launch partners and around 40 additional critical organizations, independent benchmarks can’t be established. </p><p>Examples of products launched in a similar manner that jump to mind include AI analytics tools for the intelligence and defense communities such as Palantir Gotham. </p><p>Anthropic also already provides a custom set of ‘Claude Gov’ models for US national security organizations, which have modified guardrails and are considered suitable for handling classified material.</p><p>I’ll be closely monitoring any movement in the discussions that Anthropic says it’s having with the US government on Claude Mythos Preview’s “offensive and defensive cyber capabilities”. </p><p>With Anthropic’s <a href="https://www.anthropic.com/news/where-stand-department-war" target="_blank"><u>ongoing battle with the US Department of Defense</u></a>, which has designated the frontier firm as a supply chain risk, the launch of Claude Mythos and partnership with such a wide range of tech giants could give it greater leverage over the US government.</p><p>Firms such as Microsoft, Google, Amazon, and Apple <a href="https://www.bbc.co.uk/news/articles/c4g7k7zdd0zo" target="_blank"><u>have supported Anthropic</u></a> in its legal action against the designation. All of these companies, along with other giants including Broadcom, Cisco, CrowdStrike, and Palo Alto Networks, are now publicly tied to Project Glasswing in another huge vote of confidence in Anthropic.</p><p>Google might be the most consequential partner here. Its involvement suggests that, at least for the time being, none of its Gemini models can offer the same comprehensive benefits as Claude Mythos Preview and this sends another big signal to the public sector: back this or fall behind.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/project-glasswing-anthropic-announces-big-tech-consortium-to-test-claude-mythos-ai-model-that-could-reshape-cybersecurity</link>
                                                                            <description>
                            <![CDATA[ Anthropic has announced a new initiative that will see the company collaborate with a host of tech giants to test out its Claude Mythos AI model, which the company claims could "reshape cybersecurity" ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">KYzdSgPR2Eid9wK5U8mBK4</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/YLuvjPah4So7sw69Ms6dLG-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 07 Apr 2026 18:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:12 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/YLuvjPah4So7sw69Ms6dLG-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Logo of AI company Anthropic pictured on a smartphone screen placed on top of reflective surface, with company logo reflected.]]></media:description>                                                            <media:text><![CDATA[Logo of AI company Anthropic pictured on a smartphone screen placed on top of reflective surface, with company logo reflected.]]></media:text>
                                <media:title type="plain"><![CDATA[Logo of AI company Anthropic pictured on a smartphone screen placed on top of reflective surface, with company logo reflected.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/YLuvjPah4So7sw69Ms6dLG-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Anthropic has announced the launch of a new scheme that will see the firm collaborate with an array of tech giants on the future of AI security. </p><p>Dubbed ‘Project Glasswing’, the initiative will bring together Amazon, Apple, Broadcom, Microsoft, Cisco, CrowdStrike, Palo Alto Networks, and the Linux Foundation to test a new AI model aimed at bolstering defensive cyber capabilities. </p><p>Anthropic said the model in question, Claude Mythos, has the potential to “reshape cybersecurity” and will only be available to organizations involved in the project. </p><p>As part of the scheme, launch partners will be given access to Mythos Preview for use in “defensive security work” and share learnings on the capabilities of the model. </p><p>Access has also been granted to 40 additional organisations that build or maintain critical software infrastructure in a bid to secure open source ecosystems, Anthropic revealed. </p><p>The company said it will commit up to $100 million in usage credits on this front, as well as a $4 million donation to open source security organizations. </p><h2 id="what-project-glasswing-aims-to-achieve">What Project Glasswing aims to achieve</h2><p>Project Glasswing stems from Anthropic’s own observations on the capabilities of Claude Mythos in recent weeks. Speculation over the launch of Mythos was fueled by a data leak which saw information on the general-purpose model exposed online.</p><p>That data leak pointed toward significant improvements in cybersecurity, which Anthropic has since confirmed. The company noted that it wasn’t trained specifically for <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>activities<a href="https://www.itpro.com/security/28133/what-is-cyber-security">, </a>but its capabilities are a “result of its strong agentic coding and reasoning skills”. </p><p>Simply put, Anthropic believes the model is powerful enough to warrant testing behind closed doors at trusted industry giants rather than releasing to the public, where it could be used for nefarious purposes. </p><p>“Given the rate of AI progress, it will not be long before such capabilities proliferate, potentially beyond actors who are committed to deploying them safely,” the company said in a statement.</p><p>“The work of defending the world’s cyber infrastructure might take years; frontier AI capabilities are likely to advance substantially over just the next few months. For cyber defenders to come out ahead, we need to act now.”</p><h2 id="claude-mythos-testing-shows-solid-results">Claude Mythos testing shows solid results</h2><p>According to Anthropic, Mythos Preview has identified “thousands of zero-day vulnerabilities” in recent weeks, many of which were critical. </p><p>Notably, those identified by the model were “often subtle or difficult to detect” and dating back decades. The oldest flaw detected by Mythos included a 27-year old bug in the OpenBSD operating system. </p><p>“It also discovered a 16-year-old vulnerability in a widely used video software in a line of code that automated testing tools had hit five million times without ever catching the problem,” the company said. </p><p>Anthropic revealed it is also engaged in discussions with US government officials about Claude Mythos Preview and its “offensive and defensive cyber capabilities”.</p><h2 id="a-new-cybersecurity-paradigm">A new cybersecurity paradigm</h2><p>The launch of Project Glasswing is a seismic moment for Anthropic and one that could come to define the firm as it approaches a <a href="https://seekingalpha.com/article/4887778-anthropic-ipo-what-you-need-to-know" target="_blank"><u>potential IPO</u></a> later this year.</p><p>It’s worth underlining just how unusual it is for a tech company – let alone one that’s barely five years old – to announce a gated release of its software with such widespread endorsement. </p><p>While Google and Amazon are long-time backers of Anthropic, and Microsoft has recently joined in, the full-throated support of CrowdStrike and Palo Alto Networks reads to me as recognition that Claude Mythos Preview will usher in a serious leap in cybersecurity capabilities. </p><p>Here, cybersecurity companies that have long prided themselves on their proprietary <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI </a>are admitting that Anthropic’s latest release is already catching zero days that no other tools ever have.</p><p>We’ll have to wait and see just how effective it is and, with the release limited to launch partners and around 40 additional critical organizations, independent benchmarks can’t be established. </p><p>Examples of products launched in a similar manner that jump to mind include AI analytics tools for the intelligence and defense communities such as Palantir Gotham. </p><p>Anthropic also already provides a custom set of ‘Claude Gov’ models for US national security organizations, which have modified guardrails and are considered suitable for handling classified material.</p><p>I’ll be closely monitoring any movement in the discussions that Anthropic says it’s having with the US government on Claude Mythos Preview’s “offensive and defensive cyber capabilities”. </p><p>With Anthropic’s <a href="https://www.anthropic.com/news/where-stand-department-war" target="_blank"><u>ongoing battle with the US Department of Defense</u></a>, which has designated the frontier firm as a supply chain risk, the launch of Claude Mythos and partnership with such a wide range of tech giants could give it greater leverage over the US government.</p><p>Firms such as Microsoft, Google, Amazon, and Apple <a href="https://www.bbc.co.uk/news/articles/c4g7k7zdd0zo" target="_blank"><u>have supported Anthropic</u></a> in its legal action against the designation. All of these companies, along with other giants including Broadcom, Cisco, CrowdStrike, and Palo Alto Networks, are now publicly tied to Project Glasswing in another huge vote of confidence in Anthropic.</p><p>Google might be the most consequential partner here. Its involvement suggests that, at least for the time being, none of its Gemini models can offer the same comprehensive benefits as Claude Mythos Preview and this sends another big signal to the public sector: back this or fall behind.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ AI readiness and legal compliance: Practical strategies for MSPs in the age of Copilot ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Since artificial intelligence (AI) first hit the mainstream, its capabilities have evolved significantly. AI adoption has become widespread; a <a href="https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"><u>McKinsey report</u></a> found that 62% of organizations are experimenting with AI agents to boost productivity, increase efficiency, and enable creativity across multiple businesses. </p><p>This boost in AI adoption presents a significant opportunity for Managed Service Providers (MSPs) to position themselves as trusted experts guiding customers through the AI adoption process. With proper preparation, this can help them stand out from the competition and foster long-term relationships with their clients. </p><h2 id="why-ai-is-crucial-for-msps">Why AI is crucial for MSPs</h2><p>As AI becomes commonplace, customer expectations are growing. <a href="https://www.pax8nebula.com/m/32146b2c00c945b/original/The-Agentic-Inflection-Point-Report.pdf"><u>Almost three quarters of small-to-medium businesses (SMBs) are currently experimenting with AI, with 83% of high-growth SMBs in the adoption stage</u></a>. <a href="https://www.msspalert.com/native/opentext-cybersecurity-2025-global-managed-security-survey-ai-redefines-msp-strategy"><u>Ninety-two per cent of MSPs</u></a> say that their own business has expanded as a result of an increased interest in AI. </p><p>In today’s surging market, MSPs must be prepared with an AI-ready offering to maintain their market position. Failing to do so significantly limits potential revenue and the risk of losing clients to competitors that are ahead of the curve.</p><h2 id="legal-challenges-in-the-ai-era">Legal challenges in the AI era</h2><p>Many companies are overlooking the risks of shadow AI, the unauthorized use of AI tools or applications within an organization. </p><p>The spike in interest in AI among SMBs indicates that MSPs’ customers are likely beginning to use AI tools at work.  The downside of this trend is that it raises concerns about the security of sensitive information that they neither own nor have the right to share. </p><p>An accidental confidentiality breach is often the result of human error. For example, users may be unaware of the risks of integrating personal AI tools into their work, or of the sensitive content they are dealing with.  </p><p>Businesses must ensure that privacy and HR policies are airtight and well communicated across departments.  Alongside this, the ever-changing landscape of AI regulation means legislation and guidance must be regularly reviewed to stay on track. </p><h2 id="the-benefits-and-challenges-of-copilot">The benefits and challenges of Copilot</h2><p>Designed specifically for business use, Copilot operates as an isolated instance of AI per business, enabling secure access to internal data with a reduced risk of data leaks compared to traditional OpenAI sources. </p><p>As part of the Microsoft 365 suite, Copilot ensures secure access to all data and environments, streamlining administrative tasks and saving businesses valuable time and resources.  </p><p>Despite its advantages, Copilot comes with its own security challenges. Given that it draws upon an organization’s internal data to form answers, there is always a risk that sensitive data could end up in the wrong hands. </p><p>Such breaches leave data vulnerable to being maliciously exploited, potentially leading to devastating consequences. To mitigate risk, MSPs must ensure their customers’ Microsoft 365 tenants are secured in accordance with best practices before rollout.  Microsoft data must be properly secured and strengthened access controls to prevent unauthorized users from accessing sensitive information. </p><p>Once deployed, education is paramount. MSPs should properly educate customers on how to integrate Copilot into existing processes to optimize efficiency effectively. For instance, customers should understand how Copilot integrates with the rest of their business data in Microsoft 365. </p><h2 id="how-to-become-ai-ready">How to become AI-ready</h2><p>MSPs must understand their customers’ business goals, analyse their data environment, and develop an intuitive deployment plan for Copilot ahead of selling AI services. MSPs that succeed with AI services won’t be those who rush to roll out subpar agentic enablement services. Assessing what drives business value and defining use cases is key.  </p><p>Proactively reviewing customers’ data environments to ensure that their security posture is up to scratch is critical. They should have established content management practices and data governance and carry out a thorough audit of security policies, including data access controls, retention policies, and sensitivity labels. </p><p>These are managed and updated centrally and rolled out across their Microsoft 365 environment. If security gaps are uncovered, this is a good opportunity to upsell the benefits of a Microsoft 365 Business Premium license, which provides access to Entra ID for identity access management, Purview for data security, and Defender for ransomware and device protection. </p><p>Once it’s established that the tenant is secured, MSPs can help customers to identify champions for Copilot within their organization. A small group of employees can initially test Copilot, demonstrating its potential and driving wider adoption by becoming advocates. This approach enables the organization to make the most of its 365 licenses.</p><p>Today, MSPs must not only be prepared to leverage AI internally within their operations, but also ensure they are well-equipped to support and guide customers as Copilot becomes central to their technology stack. </p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/ai-readiness-and-legal-compliance-practical-strategies-for-msps-in-the-age-of-copilot</link>
                                                                            <description>
                            <![CDATA[ How MSPs can respond effectively to the rising demand for AI services ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">kqpJvoi9BEGYnfvxCfPGTY</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/U7xFHys4ZqNqAUYxH7V7Sa-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 02 Apr 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:32:48 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Frederick Bendžius-Drennan ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/FcXiXeTL9hSHXwqugh5rsM-320-70.png ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/U7xFHys4ZqNqAUYxH7V7Sa-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Side profile view of a human head with brain synapses and flowing data lines. ]]></media:description>                                                            <media:text><![CDATA[Side profile view of a human head with brain synapses and flowing data lines. ]]></media:text>
                                <media:title type="plain"><![CDATA[Side profile view of a human head with brain synapses and flowing data lines. ]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/U7xFHys4ZqNqAUYxH7V7Sa-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Since artificial intelligence (AI) first hit the mainstream, its capabilities have evolved significantly. AI adoption has become widespread; a <a href="https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai"><u>McKinsey report</u></a> found that 62% of organizations are experimenting with AI agents to boost productivity, increase efficiency, and enable creativity across multiple businesses. </p><p>This boost in AI adoption presents a significant opportunity for Managed Service Providers (MSPs) to position themselves as trusted experts guiding customers through the AI adoption process. With proper preparation, this can help them stand out from the competition and foster long-term relationships with their clients. </p><h2 id="why-ai-is-crucial-for-msps">Why AI is crucial for MSPs</h2><p>As AI becomes commonplace, customer expectations are growing. <a href="https://www.pax8nebula.com/m/32146b2c00c945b/original/The-Agentic-Inflection-Point-Report.pdf"><u>Almost three quarters of small-to-medium businesses (SMBs) are currently experimenting with AI, with 83% of high-growth SMBs in the adoption stage</u></a>. <a href="https://www.msspalert.com/native/opentext-cybersecurity-2025-global-managed-security-survey-ai-redefines-msp-strategy"><u>Ninety-two per cent of MSPs</u></a> say that their own business has expanded as a result of an increased interest in AI. </p><p>In today’s surging market, MSPs must be prepared with an AI-ready offering to maintain their market position. Failing to do so significantly limits potential revenue and the risk of losing clients to competitors that are ahead of the curve.</p><h2 id="legal-challenges-in-the-ai-era">Legal challenges in the AI era</h2><p>Many companies are overlooking the risks of shadow AI, the unauthorized use of AI tools or applications within an organization. </p><p>The spike in interest in AI among SMBs indicates that MSPs’ customers are likely beginning to use AI tools at work.  The downside of this trend is that it raises concerns about the security of sensitive information that they neither own nor have the right to share. </p><p>An accidental confidentiality breach is often the result of human error. For example, users may be unaware of the risks of integrating personal AI tools into their work, or of the sensitive content they are dealing with.  </p><p>Businesses must ensure that privacy and HR policies are airtight and well communicated across departments.  Alongside this, the ever-changing landscape of AI regulation means legislation and guidance must be regularly reviewed to stay on track. </p><h2 id="the-benefits-and-challenges-of-copilot">The benefits and challenges of Copilot</h2><p>Designed specifically for business use, Copilot operates as an isolated instance of AI per business, enabling secure access to internal data with a reduced risk of data leaks compared to traditional OpenAI sources. </p><p>As part of the Microsoft 365 suite, Copilot ensures secure access to all data and environments, streamlining administrative tasks and saving businesses valuable time and resources.  </p><p>Despite its advantages, Copilot comes with its own security challenges. Given that it draws upon an organization’s internal data to form answers, there is always a risk that sensitive data could end up in the wrong hands. </p><p>Such breaches leave data vulnerable to being maliciously exploited, potentially leading to devastating consequences. To mitigate risk, MSPs must ensure their customers’ Microsoft 365 tenants are secured in accordance with best practices before rollout.  Microsoft data must be properly secured and strengthened access controls to prevent unauthorized users from accessing sensitive information. </p><p>Once deployed, education is paramount. MSPs should properly educate customers on how to integrate Copilot into existing processes to optimize efficiency effectively. For instance, customers should understand how Copilot integrates with the rest of their business data in Microsoft 365. </p><h2 id="how-to-become-ai-ready">How to become AI-ready</h2><p>MSPs must understand their customers’ business goals, analyse their data environment, and develop an intuitive deployment plan for Copilot ahead of selling AI services. MSPs that succeed with AI services won’t be those who rush to roll out subpar agentic enablement services. Assessing what drives business value and defining use cases is key.  </p><p>Proactively reviewing customers’ data environments to ensure that their security posture is up to scratch is critical. They should have established content management practices and data governance and carry out a thorough audit of security policies, including data access controls, retention policies, and sensitivity labels. </p><p>These are managed and updated centrally and rolled out across their Microsoft 365 environment. If security gaps are uncovered, this is a good opportunity to upsell the benefits of a Microsoft 365 Business Premium license, which provides access to Entra ID for identity access management, Purview for data security, and Defender for ransomware and device protection. </p><p>Once it’s established that the tenant is secured, MSPs can help customers to identify champions for Copilot within their organization. A small group of employees can initially test Copilot, demonstrating its potential and driving wider adoption by becoming advocates. This approach enables the organization to make the most of its 365 licenses.</p><p>Today, MSPs must not only be prepared to leverage AI internally within their operations, but also ensure they are well-equipped to support and guide customers as Copilot becomes central to their technology stack. </p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ From AI hype to AI reality: The steps businesses need to take to adopt AI responsibly ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Artificial intelligence (AI) is rapidly reshaping how organizations operate, with new models appearing almost every day. But despite the constant buzz around AI and its impact, do businesses really know how to adopt AI responsibly?</p><p>As we move past the hype phase of AI and enter what Gartner calls the ‘Trough of Disillusionment’, organizations are becoming increasingly aware of both AI’s potential value and the challenges it poses. </p><p>At the same time, business leaders are facing rising pressures to adopt sooner rather than later, as boards demand generative AI strategies and customers expect AI-augmented services. </p><p>But responsible adoption is not just a technical exercise. Instead, it is a strategic capability that requires strong foundations in governance, data, security, and workforce readiness. Without groundwork, organizations risk letting fear drive decisions, rather than a strategic approach that ensures AI delivers sustainable value.</p><h2 id="how-ai-goes-beyond-traditional-software">How AI goes beyond traditional software</h2><p>One of the most misunderstood aspects of AI readiness is that AI doesn’t behave like traditional deterministic systems. In the channel world, deploying a Power Platform automation or a standard data integration is predictable: inputs are defined, steps are repeatable, and outcomes are consistent. Apply the same configuration twice, and the result is identical.</p><p>But generative AI is not deterministic. Chatbots, Copilots, and Agents can be asked the same question multiple times and produce a different answer every time. Traditional testing and deployment approaches for digital transformation rollouts simply don’t work.</p><p>You can’t treat AI like any other software project. So, what should you do instead?</p><p>When planning AI projects, implementing responsibility and security protocols takes up the majority of a project lead’s time, rather than adopting the AI tool itself. The 40-20-40 rule is a more realistic framework to ensure AI delivers value safely, reliably, and sustainably.</p><h2 id="the-first-40-foundations-governance-and-education">The first 40% - foundations, governance, and education</h2><p>Some 40% of the time should be spent on establishing responsible data and security foundations. Before writing a single prompt, businesses need guardrails, identity management, compliance checks, secure data architecture, cleanliness, and governance. </p><p>Understanding where all data is stored and ensuring it is clean reduces the risk of AI hallucinations and improves the accuracy and reliability of outcomes. Without these measures, AI will be limited, ineffective, and potentially unsafe.</p><p>Cloud and data resilience also play a critical role. On-premises environments simply cannot deliver the scale, flexibility, or security required for modern AI systems. The cloud acts as an enabling layer for AI maturity, providing high-performance storage for large datasets and unlocking modern security and compliance capabilities. This is especially important when implementing generative AI, which interacts with sensitive data and therefore needs close governance. </p><p>A clear data strategy is equally important, as AI is only as good as the data by which it is underpinned. Structuring data appropriately, securing it correctly, and governing it responsibly will determine just how transformative AI can be. Starting with strong data and security disciplines from day one enables more powerful AI use cases, while also boosting reliability and unlocking safer automation.</p><p>AI readiness is not only a technical challenge but a cultural one. Many organisations fall into two risky extremes: employees resist AI due to fear or lack of education, or leaders adopt new tools without understanding governance, risk, or how to enable their workforce.</p><p>Neither approach delivers meaningful return on investment. To move from anxiety to empowerment, leaders must prioritize education through training, change management, and transparent communication, helping employees use AI responsibly, understand how it enhances their roles, and recognize the value it brings.</p><h2 id="the-20-implementation">The 20% - implementation</h2><p>Organizations often fixate on AI implementation. In reality, it’s the smallest slice of the effort. This phase is important, but it’s rarely where projects succeed or fail. As most businesses adopt pre-designed, embedded AI tools, such as Microsoft 365 Copilot, much of the technical heavy lifting has already been completed. </p><p>Treating implementation as the main event distracts from the deeper work required to ensure AI delivers value safely and consistently. When organisations spend too long here, it’s often a sign that earlier foundational work hasn’t been completed properly.</p><h2 id="the-final-40-optimization-resilience-and-trust">The final 40% - optimization, resilience, and trust</h2><p>The remaining 40% of the time is where AI moves from a prototype into a trustworthy business capability. This phase should focus on refining prompts, improving evaluation methods, stress-testing behaviors, and ensuring responsible AI principles are in place. </p><p>According to our research, <a href="https://www.ans.co.uk/managed-security-services/ai-readiness-secured/"><u>37% of IT decision-makers believe that cybersecurity enhancements are a challenge when adopting AI</u></a>. But strong cybersecurity testing, such as red teaming to simulate iterations on a system, is when AI transforms into a secure pillar for business success. </p><p>Red teaming simulates real-world attacks by deliberately prompting AI models and interrogating underlying data to expose vulnerabilities before they can be exploited. By involving people closest to the problem to review and critique model responses, organizations can identify weaknesses, including those revealed through malicious prompting, and strengthen both the AI system and its data.</p><p>Companies that skip this final 40% will face challenges on the road to AI adoption and are likely to struggle with accuracy, safety, and reliability. </p><h2 id="be-ready-to-evolve">Be ready to evolve</h2><p>AI adoption isn’t a one-off project. It’s an ongoing capability that will develop over time. </p><p>Organizations that succeed with AI will be those willing to evolve continuously: refining models, updating prompts, adjusting guardrails, and aligning AI initiatives with business outcomes.</p><p>Hype cycles will come and go for all innovative technologies. But readiness - across cloud, data, security, and people - is what, ultimately, turns innovation into lasting success. </p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/from-ai-hype-to-ai-reality-the-steps-businesses-need-to-take-to-adopt-ai-responsibly</link>
                                                                            <description>
                            <![CDATA[ Responsible AI adoption requires a strategic, long-term approach rather than simply deploying new tools ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">VH3Wd5LVMHUxpP6KV8dzy9</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/oEi6Z2MZjMAMQj4GokG3jc-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 31 Mar 2026 07:00:00 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:33 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Chris Huntingford ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/N3KcyjABtnSPnzdMrPW95F-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/oEi6Z2MZjMAMQj4GokG3jc-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[A light bulb on a red circuit board with a yellow digital brain inside]]></media:description>                                                            <media:text><![CDATA[A light bulb on a red circuit board with a yellow digital brain inside]]></media:text>
                                <media:title type="plain"><![CDATA[A light bulb on a red circuit board with a yellow digital brain inside]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/oEi6Z2MZjMAMQj4GokG3jc-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Artificial intelligence (AI) is rapidly reshaping how organizations operate, with new models appearing almost every day. But despite the constant buzz around AI and its impact, do businesses really know how to adopt AI responsibly?</p><p>As we move past the hype phase of AI and enter what Gartner calls the ‘Trough of Disillusionment’, organizations are becoming increasingly aware of both AI’s potential value and the challenges it poses. </p><p>At the same time, business leaders are facing rising pressures to adopt sooner rather than later, as boards demand generative AI strategies and customers expect AI-augmented services. </p><p>But responsible adoption is not just a technical exercise. Instead, it is a strategic capability that requires strong foundations in governance, data, security, and workforce readiness. Without groundwork, organizations risk letting fear drive decisions, rather than a strategic approach that ensures AI delivers sustainable value.</p><h2 id="how-ai-goes-beyond-traditional-software">How AI goes beyond traditional software</h2><p>One of the most misunderstood aspects of AI readiness is that AI doesn’t behave like traditional deterministic systems. In the channel world, deploying a Power Platform automation or a standard data integration is predictable: inputs are defined, steps are repeatable, and outcomes are consistent. Apply the same configuration twice, and the result is identical.</p><p>But generative AI is not deterministic. Chatbots, Copilots, and Agents can be asked the same question multiple times and produce a different answer every time. Traditional testing and deployment approaches for digital transformation rollouts simply don’t work.</p><p>You can’t treat AI like any other software project. So, what should you do instead?</p><p>When planning AI projects, implementing responsibility and security protocols takes up the majority of a project lead’s time, rather than adopting the AI tool itself. The 40-20-40 rule is a more realistic framework to ensure AI delivers value safely, reliably, and sustainably.</p><h2 id="the-first-40-foundations-governance-and-education">The first 40% - foundations, governance, and education</h2><p>Some 40% of the time should be spent on establishing responsible data and security foundations. Before writing a single prompt, businesses need guardrails, identity management, compliance checks, secure data architecture, cleanliness, and governance. </p><p>Understanding where all data is stored and ensuring it is clean reduces the risk of AI hallucinations and improves the accuracy and reliability of outcomes. Without these measures, AI will be limited, ineffective, and potentially unsafe.</p><p>Cloud and data resilience also play a critical role. On-premises environments simply cannot deliver the scale, flexibility, or security required for modern AI systems. The cloud acts as an enabling layer for AI maturity, providing high-performance storage for large datasets and unlocking modern security and compliance capabilities. This is especially important when implementing generative AI, which interacts with sensitive data and therefore needs close governance. </p><p>A clear data strategy is equally important, as AI is only as good as the data by which it is underpinned. Structuring data appropriately, securing it correctly, and governing it responsibly will determine just how transformative AI can be. Starting with strong data and security disciplines from day one enables more powerful AI use cases, while also boosting reliability and unlocking safer automation.</p><p>AI readiness is not only a technical challenge but a cultural one. Many organisations fall into two risky extremes: employees resist AI due to fear or lack of education, or leaders adopt new tools without understanding governance, risk, or how to enable their workforce.</p><p>Neither approach delivers meaningful return on investment. To move from anxiety to empowerment, leaders must prioritize education through training, change management, and transparent communication, helping employees use AI responsibly, understand how it enhances their roles, and recognize the value it brings.</p><h2 id="the-20-implementation">The 20% - implementation</h2><p>Organizations often fixate on AI implementation. In reality, it’s the smallest slice of the effort. This phase is important, but it’s rarely where projects succeed or fail. As most businesses adopt pre-designed, embedded AI tools, such as Microsoft 365 Copilot, much of the technical heavy lifting has already been completed. </p><p>Treating implementation as the main event distracts from the deeper work required to ensure AI delivers value safely and consistently. When organisations spend too long here, it’s often a sign that earlier foundational work hasn’t been completed properly.</p><h2 id="the-final-40-optimization-resilience-and-trust">The final 40% - optimization, resilience, and trust</h2><p>The remaining 40% of the time is where AI moves from a prototype into a trustworthy business capability. This phase should focus on refining prompts, improving evaluation methods, stress-testing behaviors, and ensuring responsible AI principles are in place. </p><p>According to our research, <a href="https://www.ans.co.uk/managed-security-services/ai-readiness-secured/"><u>37% of IT decision-makers believe that cybersecurity enhancements are a challenge when adopting AI</u></a>. But strong cybersecurity testing, such as red teaming to simulate iterations on a system, is when AI transforms into a secure pillar for business success. </p><p>Red teaming simulates real-world attacks by deliberately prompting AI models and interrogating underlying data to expose vulnerabilities before they can be exploited. By involving people closest to the problem to review and critique model responses, organizations can identify weaknesses, including those revealed through malicious prompting, and strengthen both the AI system and its data.</p><p>Companies that skip this final 40% will face challenges on the road to AI adoption and are likely to struggle with accuracy, safety, and reliability. </p><h2 id="be-ready-to-evolve">Be ready to evolve</h2><p>AI adoption isn’t a one-off project. It’s an ongoing capability that will develop over time. </p><p>Organizations that succeed with AI will be those willing to evolve continuously: refining models, updating prompts, adjusting guardrails, and aligning AI initiatives with business outcomes.</p><p>Hype cycles will come and go for all innovative technologies. But readiness - across cloud, data, security, and people - is what, ultimately, turns innovation into lasting success. </p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Systems are deterministic, people are probabilistic – AI is both, and that's a headache for cyber teams  ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Security practitioners have traditionally looked at cyber risk through a “deterministic lens”, but with AI the rules have changed and that will require a tactical rethink. </p><p>That’s according to Dean Sysman, executive chairman and co-founder of Axonius, who told attendees at the RSAC 2026 Conference that enterprise security teams need to shake up practices for an era of AI-related risks. </p><p>The technology is accelerating both offensive and defensive cyber operations globally, he noted, with bad actors now capitalizing on flaws in minutes, rather than weeks. </p><p>Similarly, unsecure AI systems mean there are more entry points for bad actors to pounce on, and enterprises are now awash with agents that have deep access to internal datasets. </p><p>With this in mind, treating AI like humans will be critical for assessing and managing risk, Sysman noted.</p><p>There is one caveat. Currently, security teams focus on deterministic behavior for IT systems and probabilistic behavior for humans. <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI</a>, however, behaves in a probabalistic <em>and</em> deterministic way.</p><p>This means managing risk is going to split their attention. </p><p>“We’ve always looked at cybersecurity through two different lenses,” he said. </p><p>“We have the deterministic lens, where we look at systems and compute and storage and memory, and those things behave in a way that we can determine, we can forecast, we can understand and react to or respond to in a deterministic way as well.</p><p>“And yet, we have people who are probabilistic. Nobody ever knows what somebody is going to do, or how they’re going to ask, or what kind of risk they’re going to introduce.”</p><p>Sysman's voice adds to a growing <a href="https://www.itpro.com/security/enterprises-need-to-think-of-agents-as-digital-co-workers-and-that-means-implementing-the-same-security-safeguards">chorus of security and IT professionals</a> calling for AI systems – and agents in particular – to be treated as human in the context of security. </p><p>A host of tech industry figures have urged enterprises to view agents as “digital co-workers” and thereby apply the same security standards. Efforts are being made to improve <a href="https://www.itpro.com/strategy/28935/what-is-identity-management-and-what-role-does-it-play-in-security-strategy">identity management</a> and observability to keep tabs on agents working away in the background.</p><p>Sysman told attendees that visibility isn’t enough, though, saying they need to shift their attention toward “actionability”. </p><p>“We need to start by seeing what we have, what exists in our environment,” he explained. “Then there’s this understanding of the organizational context of those things. We can’t fix everything and we don’t have unlimited resources, so there is always an element of prioritizing.”</p><p>Ultimately, actionability places the emphasis on ownership and accountability of AI risk in the enterprise - a key talking point during an earlier talk by Tenable co-CEO Stephen Vintz. </p><p>“The biggest question that is the hardest one to answer today [is] who owns this? Who will make this decision? Who needs to be accountable for the action and the decision that needs to be taken?”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/systems-are-deterministic-people-are-probabilistic-ai-is-both-and-thats-a-headache-for-cyber-teams</link>
                                                                            <description>
                            <![CDATA[ AI combines both the risks associated with IT systems and the people using them, creating headaches for practitioners ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">8VLzS7mBsEbtwpxLM4mgBm</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/4QoKscLUUMCfLwSt4SxwhH-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 26 Mar 2026 07:29:41 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:13 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/4QoKscLUUMCfLwSt4SxwhH-1920-80.jpg">
                                                            <media:credit><![CDATA[RSAC™ 2026 Conference]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[RSAC 2026 Conference branding pictured on a banner at the Moscone Center in San Francisco, USA.]]></media:description>                                                            <media:text><![CDATA[RSAC 2026 Conference branding pictured on a banner at the Moscone Center in San Francisco, USA.]]></media:text>
                                <media:title type="plain"><![CDATA[RSAC 2026 Conference branding pictured on a banner at the Moscone Center in San Francisco, USA.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/4QoKscLUUMCfLwSt4SxwhH-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Security practitioners have traditionally looked at cyber risk through a “deterministic lens”, but with AI the rules have changed and that will require a tactical rethink. </p><p>That’s according to Dean Sysman, executive chairman and co-founder of Axonius, who told attendees at the RSAC 2026 Conference that enterprise security teams need to shake up practices for an era of AI-related risks. </p><p>The technology is accelerating both offensive and defensive cyber operations globally, he noted, with bad actors now capitalizing on flaws in minutes, rather than weeks. </p><p>Similarly, unsecure AI systems mean there are more entry points for bad actors to pounce on, and enterprises are now awash with agents that have deep access to internal datasets. </p><p>With this in mind, treating AI like humans will be critical for assessing and managing risk, Sysman noted.</p><p>There is one caveat. Currently, security teams focus on deterministic behavior for IT systems and probabilistic behavior for humans. <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI</a>, however, behaves in a probabalistic <em>and</em> deterministic way.</p><p>This means managing risk is going to split their attention. </p><p>“We’ve always looked at cybersecurity through two different lenses,” he said. </p><p>“We have the deterministic lens, where we look at systems and compute and storage and memory, and those things behave in a way that we can determine, we can forecast, we can understand and react to or respond to in a deterministic way as well.</p><p>“And yet, we have people who are probabilistic. Nobody ever knows what somebody is going to do, or how they’re going to ask, or what kind of risk they’re going to introduce.”</p><p>Sysman's voice adds to a growing <a href="https://www.itpro.com/security/enterprises-need-to-think-of-agents-as-digital-co-workers-and-that-means-implementing-the-same-security-safeguards">chorus of security and IT professionals</a> calling for AI systems – and agents in particular – to be treated as human in the context of security. </p><p>A host of tech industry figures have urged enterprises to view agents as “digital co-workers” and thereby apply the same security standards. Efforts are being made to improve <a href="https://www.itpro.com/strategy/28935/what-is-identity-management-and-what-role-does-it-play-in-security-strategy">identity management</a> and observability to keep tabs on agents working away in the background.</p><p>Sysman told attendees that visibility isn’t enough, though, saying they need to shift their attention toward “actionability”. </p><p>“We need to start by seeing what we have, what exists in our environment,” he explained. “Then there’s this understanding of the organizational context of those things. We can’t fix everything and we don’t have unlimited resources, so there is always an element of prioritizing.”</p><p>Ultimately, actionability places the emphasis on ownership and accountability of AI risk in the enterprise - a key talking point during an earlier talk by Tenable co-CEO Stephen Vintz. </p><p>“The biggest question that is the hardest one to answer today [is] who owns this? Who will make this decision? Who needs to be accountable for the action and the decision that needs to be taken?”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Tenable co-CEO Stephen Vintz says enterprises need to get serious about tackling the AI “responsibility gap” ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Rapid <a href="https://www.itpro.com/technology/artificial-intelligence/is-enterprise-agentic-ai-adoption-matching-the-hype">AI adoption</a> is creating an array of new risks for enterprises globally, with fragmented policies on responsibility and security  exacerbating the issue. </p><p>That’s according to Tenable co-CEO, Stephen Vintz, who believes  a “responsibility gap” over <a href="https://www.itpro.com/technology/artificial-intelligence/organizations-face-ticking-timebomb-over-ai-governance">AI governance</a> has emerged over the last three years and is now one of the most pressing challenges facing organizations.</p><p>Speaking at RSAC 2026 Conference in San Francisco, Vintz told attendees that executives and boards aren’t quite sure about how to address AI-related risks, despite heavy investment in the technology. </p><p>The hype of the early <a href="https://www.itpro.com/technology/artificial-intelligence-ai/369959/what-is-generative-ai">generative AI</a> “boom” and the speed of adoption has been a key factor, according to Vintz, with enterprises rushing into adoption projects assuming long-standing governance practices will still be up to scratch. </p><p>“There is a fundamental mismatch between the exponential speed of AI adoption and the linear speed of traditional corporate governance, and things will only move faster and things will only become more complex,” he told attendees. </p><p>“Roughly 90% of all organizations have adopted AI in some form, and yet half of those have already experienced some sort of cyber incident,” Vintz added. “That’s a staggering number for a technology that’s still in its infancy.”</p><p>Establishing where responsibility for <a href="https://www.itpro.com/software/development/ai-generated-code-is-fast-becoming-the-biggest-enterprise-security-risk-as-teams-struggle-with-the-illusion-of-correctness">AI risk</a> lies is challenging because different departments are all pursuing their own goals with the technology. </p><p>Many organizations have AI-related projects spanning multiple divisions, segments, and individual teams. This results in a fragmented approach that muddles ownership, accountability, and – ultimately – security. </p><p>“The responsibility gap exists in part because <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI </a>ownership is fractured, very fractured within the enterprise,” he said. </p><p>“One of the key challenges is that AI systems are meaningfully more complex to deploy than traditional <a href="https://www.itpro.com/cloud/software-as-a-service-saas/362655/what-is-saas">SaaS </a>apps, not just technically, but also organizationally.</p><p>“The <a href="https://www.itpro.com/tag/data-science">data science</a> team owns the models, the ML Ops team owns the production pipeline, it owns deployment. The product team owns the integration of these capabilities into the offering, legal owns the compliance," he added.</p><p>From a <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>perspective, teams in this domain are then left “at the end of the line” battling to protect systems they neither designed, nor fully control. </p><h2 id="real-world-risks">Real world risks</h2><p>AI-related security incidents are not only growing in frequency, but also in scale and intensity, according to Vintz, who laid out a series of real-world examples. </p><p>In one incident, a “major financial institution” rolled out an internal AI assistant aimed at helping workers more efficiently summarize documents and speed up data analysis. </p><p>Unbeknown to the company and thousands of users, a misconfiguration meant that the agent had access to an array of sensitive information, including financial models, strategic documents, and internal communications. </p><p>“As a result, employees, thousands of them, could query it and pull information out of it that others were not meant to see,” Vintz explained. </p><p>“The exposure, well it wasn’t caused by a hacker, it was a simple misconfiguration – and who set it up? The security team, the very group responsible for protecting the organization.”</p><p>This highlights the importance of cross-functional development and risk assessment when building and deploying AI systems, Vintz said. </p><p>Each stakeholder has a crucial role to play in the process, yet the fractured approach means potentially disastrous issues or hypothetical failures are neither considered, nor tackled. </p><h2 id="global-risk">Global risk</h2><p>Ownership and accountability of AI risk isn’t just an organizational challenge at present, Vintz said, it’s a high level regulatory discussion that needs to be addressed head-on. Tools that could be employed include legislative changes, mild tweaks to existing industry frameworks, and a re-evaluation of traditional governance practices. </p><p>“To close the responsibility gap, we have to get serious about how we manage this technology and effective <a href="https://www.itpro.com/security/do-risk-awareness-and-risk-management-strategies-actually-make-a-difference">risk management</a>, and governance needs to happen in both the public sector and private sector,” he said. </p><p>With the former, Vintz said regulators will play a “massive role in setting the tone and tempo for safety”. </p><p>From an industry perspective, updates to existing security frameworks will play a key role in helping manage AI risk and setting new standards for organizations. </p><p>Vintz specifically highlighted the <a href="https://www.itpro.com/security/the-complete-guide-to-the-nist-cybersecurity-framework">NIST </a>cybersecurity framework as a “great place to start” as it’s “widely recognized as the standard for managing cyber risk”. </p><p>“It’s flexible enough to be adaptive for AI enabled threats,” he said. Similarly, at an international level <a href="https://www.itpro.com/it-governance/31712/what-is-iso-27001">ISO cybersecurity</a> standards are already being modified slightly to “bridge the gap between security and AI security”. </p><p>Ultimately, responsibility for governance will rest on the shoulders of the private sector and companies building these tools, Vintz noted. This means organizations need to start thinking about risk in a proactive way – and that will require both significant investment and a change in tack to traditional approaches. </p><p>“For over 30 years our industry has been built on firefighting, detecting … and responding at human speed,” he said. Vintz added that this made sense when risks “moved at human speed”, but that’s all changing. </p><p>“It was the right way to operate, but it fails in the AI era where everything moves at machine speed,” he said. “Our spending reflects this outdated reality. Today, over 90% of all cybersecurity spend is in detection and response.”</p><p>“Exposure management” is an approach specifically highlighted by Vintz, and one designed to tackle risks head on. </p><p>This involves a concerted focus on “unified visibility, insight, and action” which helps enterprises understand risk and then proactively reduce it. This is crucial, he noted, as risk “rarely appears in isolation”. </p><p>“Exposure management also provides the necessary intelligence layer to orchestrate the right mix of humans and AI to get stuff done, because that's really important,” Vintz said. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/tenable-co-ceo-stephen-vintz-says-enterprises-need-to-get-serious-about-tackling-the-ai-responsibility-gap</link>
                                                                            <description>
                            <![CDATA[ The Tenable chief wants a serious conversation on AI ownership and accountability ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">ty87LyLntCiNRFLDaUe8u7</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/x9sA2oW4PrD5HGM5o2k8RD-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Thu, 26 Mar 2026 07:08:20 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:13 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/x9sA2oW4PrD5HGM5o2k8RD-1920-80.jpg">
                                                            <media:credit><![CDATA[RSAC™ 2026 Conference]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[RSAC 2026 conference logos and branding pictured on screens and projected onto carpet at the Moscone Center in San Francisco, USA.]]></media:description>                                                            <media:text><![CDATA[RSAC 2026 conference logos and branding pictured on screens and projected onto carpet at the Moscone Center in San Francisco, USA.]]></media:text>
                                <media:title type="plain"><![CDATA[RSAC 2026 conference logos and branding pictured on screens and projected onto carpet at the Moscone Center in San Francisco, USA.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/x9sA2oW4PrD5HGM5o2k8RD-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Rapid <a href="https://www.itpro.com/technology/artificial-intelligence/is-enterprise-agentic-ai-adoption-matching-the-hype">AI adoption</a> is creating an array of new risks for enterprises globally, with fragmented policies on responsibility and security  exacerbating the issue. </p><p>That’s according to Tenable co-CEO, Stephen Vintz, who believes  a “responsibility gap” over <a href="https://www.itpro.com/technology/artificial-intelligence/organizations-face-ticking-timebomb-over-ai-governance">AI governance</a> has emerged over the last three years and is now one of the most pressing challenges facing organizations.</p><p>Speaking at RSAC 2026 Conference in San Francisco, Vintz told attendees that executives and boards aren’t quite sure about how to address AI-related risks, despite heavy investment in the technology. </p><p>The hype of the early <a href="https://www.itpro.com/technology/artificial-intelligence-ai/369959/what-is-generative-ai">generative AI</a> “boom” and the speed of adoption has been a key factor, according to Vintz, with enterprises rushing into adoption projects assuming long-standing governance practices will still be up to scratch. </p><p>“There is a fundamental mismatch between the exponential speed of AI adoption and the linear speed of traditional corporate governance, and things will only move faster and things will only become more complex,” he told attendees. </p><p>“Roughly 90% of all organizations have adopted AI in some form, and yet half of those have already experienced some sort of cyber incident,” Vintz added. “That’s a staggering number for a technology that’s still in its infancy.”</p><p>Establishing where responsibility for <a href="https://www.itpro.com/software/development/ai-generated-code-is-fast-becoming-the-biggest-enterprise-security-risk-as-teams-struggle-with-the-illusion-of-correctness">AI risk</a> lies is challenging because different departments are all pursuing their own goals with the technology. </p><p>Many organizations have AI-related projects spanning multiple divisions, segments, and individual teams. This results in a fragmented approach that muddles ownership, accountability, and – ultimately – security. </p><p>“The responsibility gap exists in part because <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI </a>ownership is fractured, very fractured within the enterprise,” he said. </p><p>“One of the key challenges is that AI systems are meaningfully more complex to deploy than traditional <a href="https://www.itpro.com/cloud/software-as-a-service-saas/362655/what-is-saas">SaaS </a>apps, not just technically, but also organizationally.</p><p>“The <a href="https://www.itpro.com/tag/data-science">data science</a> team owns the models, the ML Ops team owns the production pipeline, it owns deployment. The product team owns the integration of these capabilities into the offering, legal owns the compliance," he added.</p><p>From a <a href="https://www.itpro.com/security/28133/what-is-cyber-security">cybersecurity </a>perspective, teams in this domain are then left “at the end of the line” battling to protect systems they neither designed, nor fully control. </p><h2 id="real-world-risks">Real world risks</h2><p>AI-related security incidents are not only growing in frequency, but also in scale and intensity, according to Vintz, who laid out a series of real-world examples. </p><p>In one incident, a “major financial institution” rolled out an internal AI assistant aimed at helping workers more efficiently summarize documents and speed up data analysis. </p><p>Unbeknown to the company and thousands of users, a misconfiguration meant that the agent had access to an array of sensitive information, including financial models, strategic documents, and internal communications. </p><p>“As a result, employees, thousands of them, could query it and pull information out of it that others were not meant to see,” Vintz explained. </p><p>“The exposure, well it wasn’t caused by a hacker, it was a simple misconfiguration – and who set it up? The security team, the very group responsible for protecting the organization.”</p><p>This highlights the importance of cross-functional development and risk assessment when building and deploying AI systems, Vintz said. </p><p>Each stakeholder has a crucial role to play in the process, yet the fractured approach means potentially disastrous issues or hypothetical failures are neither considered, nor tackled. </p><h2 id="global-risk">Global risk</h2><p>Ownership and accountability of AI risk isn’t just an organizational challenge at present, Vintz said, it’s a high level regulatory discussion that needs to be addressed head-on. Tools that could be employed include legislative changes, mild tweaks to existing industry frameworks, and a re-evaluation of traditional governance practices. </p><p>“To close the responsibility gap, we have to get serious about how we manage this technology and effective <a href="https://www.itpro.com/security/do-risk-awareness-and-risk-management-strategies-actually-make-a-difference">risk management</a>, and governance needs to happen in both the public sector and private sector,” he said. </p><p>With the former, Vintz said regulators will play a “massive role in setting the tone and tempo for safety”. </p><p>From an industry perspective, updates to existing security frameworks will play a key role in helping manage AI risk and setting new standards for organizations. </p><p>Vintz specifically highlighted the <a href="https://www.itpro.com/security/the-complete-guide-to-the-nist-cybersecurity-framework">NIST </a>cybersecurity framework as a “great place to start” as it’s “widely recognized as the standard for managing cyber risk”. </p><p>“It’s flexible enough to be adaptive for AI enabled threats,” he said. Similarly, at an international level <a href="https://www.itpro.com/it-governance/31712/what-is-iso-27001">ISO cybersecurity</a> standards are already being modified slightly to “bridge the gap between security and AI security”. </p><p>Ultimately, responsibility for governance will rest on the shoulders of the private sector and companies building these tools, Vintz noted. This means organizations need to start thinking about risk in a proactive way – and that will require both significant investment and a change in tack to traditional approaches. </p><p>“For over 30 years our industry has been built on firefighting, detecting … and responding at human speed,” he said. Vintz added that this made sense when risks “moved at human speed”, but that’s all changing. </p><p>“It was the right way to operate, but it fails in the AI era where everything moves at machine speed,” he said. “Our spending reflects this outdated reality. Today, over 90% of all cybersecurity spend is in detection and response.”</p><p>“Exposure management” is an approach specifically highlighted by Vintz, and one designed to tackle risks head on. </p><p>This involves a concerted focus on “unified visibility, insight, and action” which helps enterprises understand risk and then proactively reduce it. This is crucial, he noted, as risk “rarely appears in isolation”. </p><p>“Exposure management also provides the necessary intelligence layer to orchestrate the right mix of humans and AI to get stuff done, because that's really important,” Vintz said. </p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Mistral CEO calls for AI cultural levy ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Major AI firms should pay a content levy in Europe, with the money going to the cultural sector, the CEO and co-founder of French AI developer Mistral has said.</p><p>Writing in the <a href="https://www.ft.com/content/d63d6291-687f-4e05-8b23-4d545d78c64a"><em>Financial Times</em></a>, Arthur Mensch said that Chinese firms are playing fast and loose with copyright rules, training their <a href="https://www.itpro.com/technology/artificial-intelligence/generative-ai-vs-large-language-models">AIs</a> on vast amounts of content – including from European sources.  </p><p>European developers, meanwhile, are being held back by a fragmented legal environment that puts them at a competitive disadvantage, he said. The current opt-out framework, which is meant to enable rights holders to protect their content and stop AI companies from using it for training if they say so, is turning out to be unworkable in practice. </p><p>"Copyrighted works continue to spread uncontrollably online, while the legal mechanisms designed to protect them remain patchy, inconsistently applied and overly complex," he said.</p><p>Mensch's solution is a revenue-based levy on all commercial providers marketing or operating AI models in Europe – including providers based abroad, for example, the US and China.</p><p>The money would go to a central European fund dedicated to investing in new content creation, as well as to supporting Europe's cultural sectors. </p><p>And, he said, the scheme would be good for <a href="https://www.itpro.com/software/development/why-python-is-the-programming-language-of-choice-for-ai-developers">AI developers</a> too, giving them legal certainty, shielding them from liability for training on materials accessible online. Importantly, it would not replace licensing agreements or the freedom to contract, he said. </p><p>"On the contrary, licensing opportunities should continue to develop and expand for usage beyond training," he said. "The fund would complement, not crowd out, direct relationships between creators and AI companies."</p><p>Mensch said he sees his proposal as a starting point, and is inviting creators, rights holders, policymakers, and fellow AI developers to get involved.</p><p>But while it does funnel money into the arts, it doesn't see individual creators remunerated for their work – something that most creators believe is crucial. </p><p>And the EU is moving in that direction. Earlier this month, the European Parliament <a href="https://www.europarl.europa.eu/news/en/press-room/20260306IPR37511/protecting-copyrighted-work-and-the-eu-s-creative-sector-in-the-age-of-ai\">adopted</a> a series of recommendations to protect copyrighted creative work, saying that EU copyright law should apply to all generative AI systems on the EU market, regardless of the place of training.</p><p>And, they said, the use of copyrighted material by generative AI must be fairly remunerated – and some way found to collect remuneration for past use. A global licence for providers to train their systems in exchange for a flat-rate payment won't do, they said.</p><p>A new licensing market for copyrighted material would involve voluntary collective licensing agreements for individual sectors, and would include individual creators and small and medium-sized enterprises. Rights holders would be able to exclude their work from being used in AI training.</p><p>"We need clear rules for the use of copyright-protected content for AI training. Legal certainty would let AI developers know which content can be used and how licences can be obtained," said rapporteur Axel Voss after the vote. </p><p>"On the other hand, rights holders would be protected against unauthorized use of their content and receive remuneration. If we want to promote and develop AI in Europe while also protecting our creators, then these provisions are absolutely indispensable."</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/mistral-ceo-calls-for-ai-cultural-levy</link>
                                                                            <description>
                            <![CDATA[ Foreign AI firms should pay for European content, says Arthur Mensch – and the European Commission is tending to agree ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">gcXKDjScSgLtjHgnydSs4d</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/fVKmsVEFVJ8unyV6qir3Mh-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 24 Mar 2026 11:30:17 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:08 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Emma Woollacott ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/aWfskavxoVSMDy6cDWtYmJ-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/fVKmsVEFVJ8unyV6qir3Mh-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[Mistral AI founder Arthur Mensch speaks at the ai-Pulse conference at Station F technology campus in Paris, France, on Friday, Nov. 17, 2023]]></media:description>                                                            <media:text><![CDATA[Mistral AI founder Arthur Mensch speaks at the ai-Pulse conference at Station F technology campus in Paris, France, on Friday, Nov. 17, 2023]]></media:text>
                                <media:title type="plain"><![CDATA[Mistral AI founder Arthur Mensch speaks at the ai-Pulse conference at Station F technology campus in Paris, France, on Friday, Nov. 17, 2023]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/fVKmsVEFVJ8unyV6qir3Mh-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Major AI firms should pay a content levy in Europe, with the money going to the cultural sector, the CEO and co-founder of French AI developer Mistral has said.</p><p>Writing in the <a href="https://www.ft.com/content/d63d6291-687f-4e05-8b23-4d545d78c64a"><em>Financial Times</em></a>, Arthur Mensch said that Chinese firms are playing fast and loose with copyright rules, training their <a href="https://www.itpro.com/technology/artificial-intelligence/generative-ai-vs-large-language-models">AIs</a> on vast amounts of content – including from European sources.  </p><p>European developers, meanwhile, are being held back by a fragmented legal environment that puts them at a competitive disadvantage, he said. The current opt-out framework, which is meant to enable rights holders to protect their content and stop AI companies from using it for training if they say so, is turning out to be unworkable in practice. </p><p>"Copyrighted works continue to spread uncontrollably online, while the legal mechanisms designed to protect them remain patchy, inconsistently applied and overly complex," he said.</p><p>Mensch's solution is a revenue-based levy on all commercial providers marketing or operating AI models in Europe – including providers based abroad, for example, the US and China.</p><p>The money would go to a central European fund dedicated to investing in new content creation, as well as to supporting Europe's cultural sectors. </p><p>And, he said, the scheme would be good for <a href="https://www.itpro.com/software/development/why-python-is-the-programming-language-of-choice-for-ai-developers">AI developers</a> too, giving them legal certainty, shielding them from liability for training on materials accessible online. Importantly, it would not replace licensing agreements or the freedom to contract, he said. </p><p>"On the contrary, licensing opportunities should continue to develop and expand for usage beyond training," he said. "The fund would complement, not crowd out, direct relationships between creators and AI companies."</p><p>Mensch said he sees his proposal as a starting point, and is inviting creators, rights holders, policymakers, and fellow AI developers to get involved.</p><p>But while it does funnel money into the arts, it doesn't see individual creators remunerated for their work – something that most creators believe is crucial. </p><p>And the EU is moving in that direction. Earlier this month, the European Parliament <a href="https://www.europarl.europa.eu/news/en/press-room/20260306IPR37511/protecting-copyrighted-work-and-the-eu-s-creative-sector-in-the-age-of-ai\">adopted</a> a series of recommendations to protect copyrighted creative work, saying that EU copyright law should apply to all generative AI systems on the EU market, regardless of the place of training.</p><p>And, they said, the use of copyrighted material by generative AI must be fairly remunerated – and some way found to collect remuneration for past use. A global licence for providers to train their systems in exchange for a flat-rate payment won't do, they said.</p><p>A new licensing market for copyrighted material would involve voluntary collective licensing agreements for individual sectors, and would include individual creators and small and medium-sized enterprises. Rights holders would be able to exclude their work from being used in AI training.</p><p>"We need clear rules for the use of copyright-protected content for AI training. Legal certainty would let AI developers know which content can be used and how licences can be obtained," said rapporteur Axel Voss after the vote. </p><p>"On the other hand, rights holders would be protected against unauthorized use of their content and receive remuneration. If we want to promote and develop AI in Europe while also protecting our creators, then these provisions are absolutely indispensable."</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Observability will be key to agentic AI safety, says Microsoft Security exec ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Enterprises around the world are flocking to agentic AI tools in 2026, with <a href="https://www.microsoft.com/en-us/security/blog/2026/02/10/80-of-fortune-500-use-active-ai-agents-observability-governance-and-security-shape-the-new-frontier/?msockid=029d9451500765ac3ba4834851a7649c" target="_blank"><u>research from Microsoft</u></a> showing that 80% of Fortune 500 companies are already using agents in daily operations. </p><p>Agents represent a step change in how enterprises leverage AI, with autonomous bots carrying out tasks on behalf of employees and helping to unlock marked productivity and efficiency gains. </p><p>Yet many organisations aren’t fully aware of the potential risks associated with these tools, according to Vasu Jakkal, corporate vice president for Microsoft Security. </p><p>Speaking during the opening keynote session at the <a href="https://www.itpro.com/uk/tag/rsac-conference">2026 RSAC Conference</a> in San Francisco, Jakkal said the integration of agents in customer-facing environments will require a re-evaluation of enterprise risk management. </p><p>Trust, Jakkal said, will be crucial for safe and secure deployment of agents, which is why observability, security, and governance will be crucial. </p><p>“Humans and agents are working together, and we are only just scratching the surface of <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI</a>, but as is always with the case of technology advancement, there will always be those who will use it for nefarious purposes,” she said, adding that the use of AI in malicious activities has reached an “inflection point”. </p><p>Microsoft’s own intelligence operations have observed bad actors using AI primarily to improve their “trade craft”. They’re using the technology to curate more efficient phishing lures and debug malware, for example. </p><p>“We’ve seen this in operations by North Korean actors Jasper Sleet [and] Coral Sleet, where AI enables sustained, large-scale misuse of legitimate access to things like identity fabrication through <a href="https://www.itpro.com/security/phishing/why-social-engineering-is-such-a-problem-and-how-your-business-can-protect-itself">social engineering</a> and really long-term persistence at very low cost.”</p><h2 id="structurally-different">“Structurally different”</h2><p>This brave new world of AI-powered malicious activity means cyber defenders now face new considerations when contending with potential risks, Jakkal said. </p><p>Indeed, malicious activities as a result of AI aren’t just faster, they’re “structurally different, and in this new reality, security has to change", Jakkal said.</p><p>Jakkal noted that organizations have relied on “layers of siloed point solutions, static policies, and human-reliant response”, but bad actors don’t take this into account. </p><p>“They think in graphs, and with agents, they can now operate continuously at machine speed across these graphs,” she explained. </p><p>This means enterprise security needs to shift from a traditional approach shoring up specific control points, toward a comprehensive architecture where defense is a proactive, not reactive approach. AI, she said, will be crucial in facilitating this change. </p><p>“At Microsoft, we believe the future of security is ambient and autonomous, just like the AI it needs to protect,” Jakkal said. </p><p>“You can’t simply turn on security, it has to be something that’s woven deeply into every layer of the AI stack – from agents to apps, to platforms, to infrastructure. It needs to be always on, always there, everywhere.</p><p>“We need to use agents. We need to use agents that are continuously discovering, testing and fixing the attack path in an always on self defending loop so defenders can address these attacks before they happen.”</p><h2 id="humans-in-the-loop">Humans in the loop</h2><p>Ensuring humans are kept in the loop will be crucial in this process and a key factor in building trust, especially given that IDC research predicts more than 1.3 billion agents will be in operation by 2028. </p><p>Areas such as identity security will become more important than ever in ensuring enterprises can keep a close eye on agents while they operate behind the scenes. </p><p>“They must be secured with the same vigilance that we use to secure people,” she said.</p><p>Similarly, the rise of “double agents” – those that have been manipulated by malicious actors to engage in nefarious activities – have already been observed by Microsoft. </p><p>With this in mind, Jakkal expects observability to be a key enterprise focus in the coming years. </p><p>“We cannot protect what we cannot see,” she said. “And in this era of agentic AI, organizations will need an observability control plane.”</p><p>Observability won’t rest solely with security teams either, she said. Developer teams and IT teams will also require shared controls to shore up identity and data security, and to ensure robust governance of agents. </p><p>The stakes are high when it comes to safe and secure agentic AI adoption, Jakkal said, which underlines the need for a trustworthy approach to integration. It will also have long-term positive implications for enterprises, if done correctly. </p><p>“As we do this, I know that we will build trust at the very core of our organizations, and security becomes that incredible catalyst for innovation.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/observability-will-be-key-to-agentic-ai-safety-says-microsoft-security-exec</link>
                                                                            <description>
                            <![CDATA[ Agentic AI adoption will require a re-evaluation of enterprise risk management, according to Microsoft corporate VP ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">wEcVtXavZQbHprcRW7rPtk</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/Ux3V79nc6iXyfwiHXRzQi3-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 24 Mar 2026 06:55:47 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:23 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/Ux3V79nc6iXyfwiHXRzQi3-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[AI zero trust concept image showing shield symbol with digitized human brain in a circle, with distributed red data points with skull symbols.]]></media:description>                                                            <media:text><![CDATA[AI zero trust concept image showing shield symbol with digitized human brain in a circle, with distributed red data points with skull symbols.]]></media:text>
                                <media:title type="plain"><![CDATA[AI zero trust concept image showing shield symbol with digitized human brain in a circle, with distributed red data points with skull symbols.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/Ux3V79nc6iXyfwiHXRzQi3-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Enterprises around the world are flocking to agentic AI tools in 2026, with <a href="https://www.microsoft.com/en-us/security/blog/2026/02/10/80-of-fortune-500-use-active-ai-agents-observability-governance-and-security-shape-the-new-frontier/?msockid=029d9451500765ac3ba4834851a7649c" target="_blank"><u>research from Microsoft</u></a> showing that 80% of Fortune 500 companies are already using agents in daily operations. </p><p>Agents represent a step change in how enterprises leverage AI, with autonomous bots carrying out tasks on behalf of employees and helping to unlock marked productivity and efficiency gains. </p><p>Yet many organisations aren’t fully aware of the potential risks associated with these tools, according to Vasu Jakkal, corporate vice president for Microsoft Security. </p><p>Speaking during the opening keynote session at the <a href="https://www.itpro.com/uk/tag/rsac-conference">2026 RSAC Conference</a> in San Francisco, Jakkal said the integration of agents in customer-facing environments will require a re-evaluation of enterprise risk management. </p><p>Trust, Jakkal said, will be crucial for safe and secure deployment of agents, which is why observability, security, and governance will be crucial. </p><p>“Humans and agents are working together, and we are only just scratching the surface of <a href="https://www.itpro.com/strategy/28181/what-is-ai">AI</a>, but as is always with the case of technology advancement, there will always be those who will use it for nefarious purposes,” she said, adding that the use of AI in malicious activities has reached an “inflection point”. </p><p>Microsoft’s own intelligence operations have observed bad actors using AI primarily to improve their “trade craft”. They’re using the technology to curate more efficient phishing lures and debug malware, for example. </p><p>“We’ve seen this in operations by North Korean actors Jasper Sleet [and] Coral Sleet, where AI enables sustained, large-scale misuse of legitimate access to things like identity fabrication through <a href="https://www.itpro.com/security/phishing/why-social-engineering-is-such-a-problem-and-how-your-business-can-protect-itself">social engineering</a> and really long-term persistence at very low cost.”</p><h2 id="structurally-different">“Structurally different”</h2><p>This brave new world of AI-powered malicious activity means cyber defenders now face new considerations when contending with potential risks, Jakkal said. </p><p>Indeed, malicious activities as a result of AI aren’t just faster, they’re “structurally different, and in this new reality, security has to change", Jakkal said.</p><p>Jakkal noted that organizations have relied on “layers of siloed point solutions, static policies, and human-reliant response”, but bad actors don’t take this into account. </p><p>“They think in graphs, and with agents, they can now operate continuously at machine speed across these graphs,” she explained. </p><p>This means enterprise security needs to shift from a traditional approach shoring up specific control points, toward a comprehensive architecture where defense is a proactive, not reactive approach. AI, she said, will be crucial in facilitating this change. </p><p>“At Microsoft, we believe the future of security is ambient and autonomous, just like the AI it needs to protect,” Jakkal said. </p><p>“You can’t simply turn on security, it has to be something that’s woven deeply into every layer of the AI stack – from agents to apps, to platforms, to infrastructure. It needs to be always on, always there, everywhere.</p><p>“We need to use agents. We need to use agents that are continuously discovering, testing and fixing the attack path in an always on self defending loop so defenders can address these attacks before they happen.”</p><h2 id="humans-in-the-loop">Humans in the loop</h2><p>Ensuring humans are kept in the loop will be crucial in this process and a key factor in building trust, especially given that IDC research predicts more than 1.3 billion agents will be in operation by 2028. </p><p>Areas such as identity security will become more important than ever in ensuring enterprises can keep a close eye on agents while they operate behind the scenes. </p><p>“They must be secured with the same vigilance that we use to secure people,” she said.</p><p>Similarly, the rise of “double agents” – those that have been manipulated by malicious actors to engage in nefarious activities – have already been observed by Microsoft. </p><p>With this in mind, Jakkal expects observability to be a key enterprise focus in the coming years. </p><p>“We cannot protect what we cannot see,” she said. “And in this era of agentic AI, organizations will need an observability control plane.”</p><p>Observability won’t rest solely with security teams either, she said. Developer teams and IT teams will also require shared controls to shore up identity and data security, and to ensure robust governance of agents. </p><p>The stakes are high when it comes to safe and secure agentic AI adoption, Jakkal said, which underlines the need for a trustworthy approach to integration. It will also have long-term positive implications for enterprises, if done correctly. </p><p>“As we do this, I know that we will build trust at the very core of our organizations, and security becomes that incredible catalyst for innovation.”</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Safe AI adoption rests on cybersecurity professionals, says RSAC chairman ]]></title>
                                                                                                <dc:content><![CDATA[ <p><a href="https://www.itpro.com/security/28133/what-is-cyber-security">Cybersecurity </a>professionals are in the vanguard of driving safe, responsible AI adoption, according to RSAC executive chairman Hugh Thompson. </p><p>Speaking during the opening keynote of <a href="https://www.itpro.com/tag/rsac-conference">RSAC Conference 2026</a> in San Francisco, Thompson said security professionals have never been more important amidst a period of growing cyber risks and rapid technological change. </p><p>“There’s a certain dynamism that exists right now in cybersecurity,” he said. “AI is rapidly changing everything around us. It’s changing everything, it’s forcing us to rethink how we [keep information safe].”</p><p><a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> are helping security practitioners respond to threats more efficiently than ever, Thompson noted, unlocking significant productivity gains. </p><p>Yet while the benefits of the technology are becoming clearer in cyber, malicious actors are also flocking to these tools. Indeed, Thompson said the increased use of AI among bad actors is forcing enterprises to “reimagine” how these tools can be used for nefarious purposes. </p><p>The duality of AI has been a recurring talking point over the last two years. The technology is delivering benefits for organizations and individual workers, yet a host of studies over the last year have highlighted the growing use of AI among malicious actors. </p><p>Recent <a href="https://www.itpro.com/security/phishing/ai-generated-phishing-became-the-baseline-for-hackers-last-year-kaseya-warns-its-going-to-get-worse-in-2026"><u>analysis from Kaseya</u></a>, for example, showed 2025 was an “inflection point” for AI and cybersecurity, with traditional tactics such as phishing now supercharged by the use of the technology. </p><p>Around 83% of <a href="https://www.itpro.com/security/29093/what-is-phishing">phishing </a>emails use AI content in some way, the study found, with 40% of <a href="https://www.itpro.com/security/cyber-attacks/what-is-business-email-compromise-bec">business email compromise (BEC)</a> techniques now using generative AI.</p><p>This is where the cybersecurity function will be a critical factor in ensuring safe adoption of the technology en-masse globally. Moreover, cyber professionals have an opportunity to take the lead in facilitating this mass shift. </p><p>“We cannot be passive observers on this AI journey,” Thompson told attendees. “AI and cybersecurity are so deeply intertwined, we can’t let AI be something that happens to us. Instead, it’s our responsibility as cybersecurity professionals to make AI work for us.”</p><h2 id="ai-needs-cyber-pros">AI needs cyber pros</h2><p>Thompson noted that AI “cannot operate sustainably” without robust security safeguards and practitioners offering a guiding hand for IT leaders. </p><p>With that in mind, fostering a more synergetic relationship between this domain and other business functions will be crucial – and RSAC Conference represents an environment for professionals to engage on the topic. </p><p>“I would argue AI just made our jobs way bigger in cybersecurity,” he said. “Conversations that will happen this week, in the sessions, in the halls, will help shape the future of AI for years to come.”</p><p>The power of community was a another key talking point throughout Thompson’s opening keynote, which saw him joined on stage by newly-appointed RSAC chief executive, Jen Easterly. </p><p>Easterly said the strength of the global cybersecurity community will stand organisations in good stead during a period of technological change in which malicious actors are becoming increasingly aggressive. </p><p>“This is our community, and we should draw strength from that, because together, we are stronger than any threat,” she said. </p><p>“Together, we're building trust in a world that desperately needs trust, a world increasingly powered by the most consequential technology of our lifetime, moving faster and faster than ever.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/security/safe-ai-adoption-rests-on-cybersecurity-professionals-says-rsac-chairman</link>
                                                                            <description>
                            <![CDATA[ With AI security a key talking point at RSAC 2026, executive chairman Hugh Thompson believes the industry can lead by example ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">N5ZmUJcHTmNAsbVhUNW9sF</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/3JmgCnSE8HxZ5ARAA9imGG-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Tue, 24 Mar 2026 06:22:53 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:13 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Security]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ ross.kelly@futurenet.com (Ross Kelly) ]]></author>                    <dc:creator><![CDATA[ Ross Kelly ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Y5vrV2V98Np6jHAGmAtCd3-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Ross Kelly is ITPro&#039;s News &amp;amp; Analysis Editor, with a keen interest in cyber security, business leadership and emerging technologies.&lt;/p&gt;
&lt;p&gt;He graduated from Edinburgh Napier University in 2016 with a BA (Hons) in Journalism, and joined ITPro in 2022 after four years working in technology conference research.&amp;nbsp;&lt;/p&gt;
&lt;p&gt;In his spare time, Ross enjoys cycling, walking and is an avid reader of history and non-fiction.&lt;/p&gt;
&lt;p&gt;You can contact Ross at ross.kelly@futurenet.com or on &lt;a href=&quot;https://twitter.com/rosswritesetc&quot;&gt;Twitter&lt;/a&gt; and &lt;a href=&quot;https://www.linkedin.com/in/ross-kelly-18a54411a/&quot;&gt;LinkedIn&lt;/a&gt;.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/3JmgCnSE8HxZ5ARAA9imGG-1920-80.jpg">
                                                            <media:credit><![CDATA[RSAC]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[RSAC executive chairman Hugh Thompson pictured on stage during the opening keynote at the 2026 RSAC Conference in San Francisco, USA.]]></media:description>                                                            <media:text><![CDATA[RSAC executive chairman Hugh Thompson pictured on stage during the opening keynote at the 2026 RSAC Conference in San Francisco, USA.]]></media:text>
                                <media:title type="plain"><![CDATA[RSAC executive chairman Hugh Thompson pictured on stage during the opening keynote at the 2026 RSAC Conference in San Francisco, USA.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/3JmgCnSE8HxZ5ARAA9imGG-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p><a href="https://www.itpro.com/security/28133/what-is-cyber-security">Cybersecurity </a>professionals are in the vanguard of driving safe, responsible AI adoption, according to RSAC executive chairman Hugh Thompson. </p><p>Speaking during the opening keynote of <a href="https://www.itpro.com/tag/rsac-conference">RSAC Conference 2026</a> in San Francisco, Thompson said security professionals have never been more important amidst a period of growing cyber risks and rapid technological change. </p><p>“There’s a certain dynamism that exists right now in cybersecurity,” he said. “AI is rapidly changing everything around us. It’s changing everything, it’s forcing us to rethink how we [keep information safe].”</p><p><a href="https://www.itpro.com/technology/artificial-intelligence/amazing-ai-tools-to-try-today">AI tools</a> are helping security practitioners respond to threats more efficiently than ever, Thompson noted, unlocking significant productivity gains. </p><p>Yet while the benefits of the technology are becoming clearer in cyber, malicious actors are also flocking to these tools. Indeed, Thompson said the increased use of AI among bad actors is forcing enterprises to “reimagine” how these tools can be used for nefarious purposes. </p><p>The duality of AI has been a recurring talking point over the last two years. The technology is delivering benefits for organizations and individual workers, yet a host of studies over the last year have highlighted the growing use of AI among malicious actors. </p><p>Recent <a href="https://www.itpro.com/security/phishing/ai-generated-phishing-became-the-baseline-for-hackers-last-year-kaseya-warns-its-going-to-get-worse-in-2026"><u>analysis from Kaseya</u></a>, for example, showed 2025 was an “inflection point” for AI and cybersecurity, with traditional tactics such as phishing now supercharged by the use of the technology. </p><p>Around 83% of <a href="https://www.itpro.com/security/29093/what-is-phishing">phishing </a>emails use AI content in some way, the study found, with 40% of <a href="https://www.itpro.com/security/cyber-attacks/what-is-business-email-compromise-bec">business email compromise (BEC)</a> techniques now using generative AI.</p><p>This is where the cybersecurity function will be a critical factor in ensuring safe adoption of the technology en-masse globally. Moreover, cyber professionals have an opportunity to take the lead in facilitating this mass shift. </p><p>“We cannot be passive observers on this AI journey,” Thompson told attendees. “AI and cybersecurity are so deeply intertwined, we can’t let AI be something that happens to us. Instead, it’s our responsibility as cybersecurity professionals to make AI work for us.”</p><h2 id="ai-needs-cyber-pros">AI needs cyber pros</h2><p>Thompson noted that AI “cannot operate sustainably” without robust security safeguards and practitioners offering a guiding hand for IT leaders. </p><p>With that in mind, fostering a more synergetic relationship between this domain and other business functions will be crucial – and RSAC Conference represents an environment for professionals to engage on the topic. </p><p>“I would argue AI just made our jobs way bigger in cybersecurity,” he said. “Conversations that will happen this week, in the sessions, in the halls, will help shape the future of AI for years to come.”</p><p>The power of community was a another key talking point throughout Thompson’s opening keynote, which saw him joined on stage by newly-appointed RSAC chief executive, Jen Easterly. </p><p>Easterly said the strength of the global cybersecurity community will stand organisations in good stead during a period of technological change in which malicious actors are becoming increasingly aggressive. </p><p>“This is our community, and we should draw strength from that, because together, we are stronger than any threat,” she said. </p><p>“Together, we're building trust in a world that desperately needs trust, a world increasingly powered by the most consequential technology of our lifetime, moving faster and faster than ever.</p><h3 class="article-body__section" id="section-follow-us-on-social-media"><span>FOLLOW US ON SOCIAL MEDIA</span></h3>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ Will AI hiring entrench gender bias? ]]></title>
                                                                                                <dc:content><![CDATA[ <iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/1b6bf04c-ed92-440f-9d96-fcf042128d42/" src="https://player.captivate.fm/episode/1b6bf04c-ed92-440f-9d96-fcf042128d42/"></iframe><p>Women are still greatly underrepresented in STEM. The latest figures from the Women Tech Network show women only make up around 26-27% of the STEM workforce and the organization estimates that at the current rate of change, it will be nearly 123 years before the economic gender gap is closed.</p><p>It’s International Women’s Day 2026 – and the rate of women hired in tech continues to lag far behind that of men.</p><p> As hiring teams turn to AI tools to automatically field candidates, we’re at something of a crossroads – do we fix the bias inherent in hiring? Or simply replicate it at scale with similarly biased AI tools?</p><p>In this episode, Jane are Rory are joined by Clare Hickie, EMEA CTO at Workday, to discuss how businesses can engage in bias-free talent acquisition in the age of AI.</p><h2 id="highlights-4">Highlights</h2><p>"It's not a problem if you control it, and the only way that you can control it is to ensure that you have got a responsible AI program, the right measures, the right procedures, the right guardrails and the right protocols in place."</p><p>"I've been in technology a long, long time, and it's now it's my job to really pay back. And my job in paying back is by ensuring that I can go out and really encourage and inspire young people into technology. So there's a couple of things going on."</p><p>"We've seen a negative impact on society as a consequence of AI as well, right? But what we're really are driving for is that positive impact on society, and the only way you can do that is removing the bias of the past. And the way you remove that biased information is to use the technology to also ensuring that you're operating with clean data and with context.</p><p>"You have to apply fair processes. You have to apply, apply fair procedures. So in the past, organizations, for example, didn't even have to post that there was another job, it was a tap on the shoulder after another tap on the shoulder. And sometimes, you know what? It worked and then for many they were just not even considered, because they were just forming these groups and teams and leadership teams that they were all from themselves.</p><h2 id="footnotes-2">Footnotes</h2><ul><li><a href="https://www.itpro.com/security/why-are-there-still-so-few-women-working-in-cybersecurity">International Women’s Day 2026: why are there still so few women working in cybersecurity?</a></li><li><a href="https://www.itpro.com/business/careers-and-training/gender-diversity-improvements-could-be-the-key-to-tackling-the-uks-ai-skills-shortage">Gender diversity improvements could be the key to tackling the UK's AI skills shortage</a></li><li><a href="https://www.itpro.com/business/careers-and-training/why-are-so-many-men-in-tech-blind-to-the-gender-divide">Why are so many men in tech blind to the gender divide?</a></li><li><a href="https://www.itpro.com/business/leadership/is-diversity-still-a-challenge-in-the-channel">Is diversity still a challenge in the channel?</a></li></ul><h2 id="subscribe-2">Subscribe </h2><ul><li><a href="https://apple.sjv.io/c/221109/473657/7613?subId1=itpro-gb-1243831151189624600&amp;sharedId=itpro-gb&amp;u=https%3A%2F%2Fpodcasts.apple.com%2Fgb%2Fpodcast%2Fthe-itpro-podcast%2Fid1483810154" target="_blank"><u>Subscribe to The IT Pro Podcast on Apple Podcasts</u></a></li><li><a href="https://open.spotify.com/show/7HpYehTy752KmtbwpOAgRZ" target="_blank"><u>Subscribe to The IT Pro Podcast on Spotify</u></a></li><li><a href="https://www.itpro.co.uk/newsletter-signup"><u>Subscribe to the IT Pro newsletter</u></a></li><li><a href="https://uk.linkedin.com/company/itpro-uk" target="_blank"><u>Join us on LinkedIn</u></a></li></ul> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/will-ai-hiring-entrench-gender-bias</link>
                                                                            <description>
                            <![CDATA[ This International Women's Day, it's more important than ever to consider the inherent biases of training data ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">4N46x8MECbhxeHggL8kuiH</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/QzJ9ZNCnFifc3yX4DWnaBD-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 06 Mar 2026 13:43:50 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:22 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                <author><![CDATA[ rory.bathgate@futurenet.com (Rory Bathgate) ]]></author>                    <dc:creator><![CDATA[ Rory Bathgate ]]></dc:creator>                                                                                    <dc:source><![CDATA[ https://cdn.mos.cms.futurecdn.net/Vep5JogbPhduK7R6CUWAm6-320-70.jpg ]]></dc:source>
                                                                <dc:description><![CDATA[ &lt;p&gt;Rory Bathgate was the Features and Multimedia Editor at ITPro until 2026, overseeing all in-depth content and case studies. Throughout his time at ITPro, Rory  charted the rise in popularity of generative AI and specifically companies such as Microsoft, OpenAI, and Google. &lt;/p&gt;&lt;p&gt;Alongside this, he delved into increasing calls for ethical and responsible AI as global legislators circle the technology, as well as the latest in mobile networking technology, from 5G mmWave to the 3G sunset and how it will affect businesses.&lt;/p&gt;&lt;p&gt;He  provided coverage from high-profile tech conferences such as Dell Technologies World, SuiteWorld, and VMware Explore Europe. His on-the-ground coverage  included live blogs, extensive daily coverage of the most significant announcements, analysis pieces, and podcast.&lt;/p&gt;&lt;p&gt;Rory was the co-host of the ITPro Podcast alongside Jane McCallion, where he swapped a keyboard for a microphone to discuss the latest learnings in tech.&lt;/p&gt;&lt;p&gt;Rory has an MA in Eighteenth-Century Studies from King’s College London, as well as a BA in English and American Literature from the University of Kent. He joined ITPro in 2022 as a graduate, after four years in student journalism.&lt;/p&gt;&lt;p&gt;In his free time, Rory enjoys photography and video editing, and can often be found at the cinema or reading a good science fiction paperback.&lt;/p&gt; ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/QzJ9ZNCnFifc3yX4DWnaBD-1920-80.jpg">
                                                            <media:credit><![CDATA[Future]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[The text &quot;Will AI hiring entrench gender bias?&quot; against a black and white pie chart on top of a purple background. The words &quot;AI hiring&quot; and &quot;gender bias&quot; are in yellow, the rest are in white. In the bottom-right corner, the ITPro Podcast logo is shown.]]></media:description>                                                            <media:text><![CDATA[The text &quot;Will AI hiring entrench gender bias?&quot; against a black and white pie chart on top of a purple background. The words &quot;AI hiring&quot; and &quot;gender bias&quot; are in yellow, the rest are in white. In the bottom-right corner, the ITPro Podcast logo is shown.]]></media:text>
                                <media:title type="plain"><![CDATA[The text &quot;Will AI hiring entrench gender bias?&quot; against a black and white pie chart on top of a purple background. The words &quot;AI hiring&quot; and &quot;gender bias&quot; are in yellow, the rest are in white. In the bottom-right corner, the ITPro Podcast logo is shown.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/QzJ9ZNCnFifc3yX4DWnaBD-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/1b6bf04c-ed92-440f-9d96-fcf042128d42/" src="https://player.captivate.fm/episode/1b6bf04c-ed92-440f-9d96-fcf042128d42/"></iframe><p>Women are still greatly underrepresented in STEM. The latest figures from the Women Tech Network show women only make up around 26-27% of the STEM workforce and the organization estimates that at the current rate of change, it will be nearly 123 years before the economic gender gap is closed.</p><p>It’s International Women’s Day 2026 – and the rate of women hired in tech continues to lag far behind that of men.</p><p> As hiring teams turn to AI tools to automatically field candidates, we’re at something of a crossroads – do we fix the bias inherent in hiring? Or simply replicate it at scale with similarly biased AI tools?</p><p>In this episode, Jane are Rory are joined by Clare Hickie, EMEA CTO at Workday, to discuss how businesses can engage in bias-free talent acquisition in the age of AI.</p><h2 id="highlights-4">Highlights</h2><p>"It's not a problem if you control it, and the only way that you can control it is to ensure that you have got a responsible AI program, the right measures, the right procedures, the right guardrails and the right protocols in place."</p><p>"I've been in technology a long, long time, and it's now it's my job to really pay back. And my job in paying back is by ensuring that I can go out and really encourage and inspire young people into technology. So there's a couple of things going on."</p><p>"We've seen a negative impact on society as a consequence of AI as well, right? But what we're really are driving for is that positive impact on society, and the only way you can do that is removing the bias of the past. And the way you remove that biased information is to use the technology to also ensuring that you're operating with clean data and with context.</p><p>"You have to apply fair processes. You have to apply, apply fair procedures. So in the past, organizations, for example, didn't even have to post that there was another job, it was a tap on the shoulder after another tap on the shoulder. And sometimes, you know what? It worked and then for many they were just not even considered, because they were just forming these groups and teams and leadership teams that they were all from themselves.</p><h2 id="footnotes-2">Footnotes</h2><ul><li><a href="https://www.itpro.com/security/why-are-there-still-so-few-women-working-in-cybersecurity">International Women’s Day 2026: why are there still so few women working in cybersecurity?</a></li><li><a href="https://www.itpro.com/business/careers-and-training/gender-diversity-improvements-could-be-the-key-to-tackling-the-uks-ai-skills-shortage">Gender diversity improvements could be the key to tackling the UK's AI skills shortage</a></li><li><a href="https://www.itpro.com/business/careers-and-training/why-are-so-many-men-in-tech-blind-to-the-gender-divide">Why are so many men in tech blind to the gender divide?</a></li><li><a href="https://www.itpro.com/business/leadership/is-diversity-still-a-challenge-in-the-channel">Is diversity still a challenge in the channel?</a></li></ul><h2 id="subscribe-2">Subscribe </h2><ul><li><a href="https://apple.sjv.io/c/221109/473657/7613?subId1=itpro-gb-1243831151189624600&amp;sharedId=itpro-gb&amp;u=https%3A%2F%2Fpodcasts.apple.com%2Fgb%2Fpodcast%2Fthe-itpro-podcast%2Fid1483810154" target="_blank"><u>Subscribe to The IT Pro Podcast on Apple Podcasts</u></a></li><li><a href="https://open.spotify.com/show/7HpYehTy752KmtbwpOAgRZ" target="_blank"><u>Subscribe to The IT Pro Podcast on Spotify</u></a></li><li><a href="https://www.itpro.co.uk/newsletter-signup"><u>Subscribe to the IT Pro newsletter</u></a></li><li><a href="https://uk.linkedin.com/company/itpro-uk" target="_blank"><u>Join us on LinkedIn</u></a></li></ul>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
                                <item>
                                                            <title><![CDATA[ In from the shadows: AI’s uncontrolled growth ]]></title>
                                                                                                <dc:content><![CDATA[ <p>Artificial intelligence is one of the fastest-growing technologies of all time.</p><p>The United Nations’ trade and development agency, UNCTAD, <a href="https://unctad.org/news/ai-market-projected-hit-48-trillion-2033-emerging-dominant-frontier-technology"><u>predicts</u></a> that the global AI market will be worth$4.8 trillion by 2033, or 25 times its value just three years ago. </p><p><a href="https://www.itpro.com/technology/artificial-intelligence/openai-just-revealed-what-people-really-use-chatgpt-for-and-70-percent-of-queries-have-nothing-to-do-with-work">ChatGPT</a>, which popularised generative AI and <a href="https://www.itpro.com/technology/artificial-intelligence/generative-ai-vs-large-language-models">large language models (LLMs)</a>, reached 100m active users within just two months of its launch in 2022, making it the fastest-growing consumer app ever.</p><p>As ChatGPT’s growth suggests, however, the initial uptake of generative AI has been largely outside corporate IT, with CIOs and enterprises playing catch up ever since.</p><p>Fast-growing, free-to-use large language models and generative AI tools have created a new layer of technology, largely outside the control of corporate IT. This so-called shadow AI has grown, even as enterprise software vendors have added AI functions to their applications.</p><p>While <a href="https://www.itpro.com/technology/artificial-intelligence/gartner-says-40-percent-of-enterprises-will-experience-shadow-ai-breaches-by-2030-educating-staff-is-the-key-to-avoiding-disaster"><u>shadow AI</u></a> has parallels to the shadow IT challenges of the last two decades, uncontrolled use of AI is bringing its own, significant challenges.</p><p>These include breaches of privacy and confidentiality, loss of intellectual property, security issues, and even incorrect and damaging decisions made by shadow AI.</p><h2 id="hiding-in-plain-sight">Hiding in plain sight</h2><p><a href="https://www.gartner.com/en/newsroom/press-releases/2025-11-19-gartner-identifies-critical-genai-blind-spots-that-cios-must-urgently-address0" target="_blank"><u>According to industry analysts Gartner</u></a>, IT leaders in 69% of organizations suspect, or know, that staff are using “unauthorised AI”. The firm also predicts that 40% of businesses will see “security or compliance incidents” resulting from unauthorised AI. </p><p>Security and privacy experts are increasingly worried about the downsides of unauthorised AI, especially as it spreads from simple chatbots to more critical areas such as code development (vibe coding), or <a href="https://www.itpro.com/technology/artificial-intelligence/how-ai-agent-boss-is-reshaping-it-accountability"><u>“agentic” systems</u></a> that operate with little in the way of human oversight.</p><p>Current IT and data security systems are not well-placed to detect and block AI tools experts warn.</p><p>“AI has been available for about three years now, and as such, people are getting used to which platforms they get good results from,” James Gillies, head of cyber security at Logicalis UKI, tells <em>ITPro</em>.</p><p>“It’s become ‘easy’ to get a quick answer to almost any question, but oftentimes the consideration of asking the question in the first place gets forgotten, or worse still, ignored,” he warns.</p><p>This sees employees putting confidential and sensitive information into AI tools, with little or no consideration about where that data goes and how it might be used or reused.</p><p>Employees will continue to use shadow AI, unless CIOs and CISOs can implement strict control measures. Even then, enterprises need to balance the dangers of AI, with the risk of stifling innovation.</p><p>“Unlike traditional shadow IT, which typically involves unauthorised apps or cloud services, shadow AI introduces unique risks because of the way AI tools operate and interact with sensitive data,” warns Findlay Whitelaw, a cybersecurity strategist and researcher at security company Exabeam.</p><p>“Information shared with AI tools may be stored, reused or transferred across jurisdictions often outside organizational visibility or control,” she tells <em>ITPro</em>. “What adds to the complexity of shadow AI is just how widespread its consequences can be.”</p><iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/" src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/"></iframe><h2 id="inside-the-machine">Inside the machine</h2><p>For CIOs and those in the enterprise tasked with data security and privacy, there is an urgent need to break down the problem posed by shadow AI. This is not easy.</p><p>“Shadow AI is a bigger concern than shadow IT as it has identity, access and infrastructure impact,” Jon Collins, field CTO at analysts GigaOm, tells ITPro.</p><p>“Shadow IT came with risks: accessing corporate data on unsecured personal devices for example, or using Dropbox and the like for file exchange with data leakage potential, or SaaS systems for project and data management. Each had an operational overhead and was – and remains – a source of pain,” he says.</p><p>“Shadow AI, when confined to the LLM or chatbot world, has similar risks. Additional risks come from agentic approaches, which operate autonomously and at scale. [It’s] no longer a single person causing issues or creating risk, but a single person, potentially inadvertently, launching a flotilla of uncontrolled, autonomous operators into an infrastructure environment which has not been set up to protect against them.</p><p>“Our internal infrastructures have gaping holes of misconfigured systems, poor deprovisioning and access practices, open APIs and mismanaged secrets, which agents can catalogue, navigate and exploit.”</p><p>As Collins describes it, many of the risks from shadow AI are similar to those posed by any unauthorised and uncontrolled technology. For CIOs, this is not a new challenge. Adopting consumer IT in the enterprise has always been a trade off between control and usability, but AI amplifies the problem, both through its complexity and because humans tend to trust machines.</p><h2 id="issues-of-trust">Issues of trust</h2><p>“Shadow AI often exploits what is often described as ‘machine trust’, the assumption that algorithms are inherently accurate and reliable,” suggests Exabeam’s Whitelaw. “Despite this default trust, AI agents are only as good as the data they’re trained on and can carry biases, errors, or vulnerabilities that can be inadvertently, or even intentionally, exploited.”</p><p>Even where individual users trust personal AI tools and perhaps prefer them to corporate technology, they are unlikely to have insights into AI’s algorithms, or details of the data vendors use to train models.</p><p>This, in turn, can lead to unexpected and unwelcome results and the enterprise can be held liable. Controlling shadow AI needs monitoring tools, mapping AI tools to data and business risk, and workforce education.</p><p>“Organizations will have to look into the usage of AI models. To get more visibility into the models they use, they will have to manage usage and make sure they’re not introducing malicious models into their infrastructure,” Ron Minis, senior security researcher at JFrog, tells <em>ITPro</em>.</p><p>However, <a href="https://www.itpro.com/technology/artificial-intelligence/shadow-ai-can-be-a-tool-for-ai-innovation-with-the-right-controls-say-gartner-analysts"><u>banning AI won’t work</u></a>, says Minis. “If companies think they can block or somehow prevent the usage of AI In the company, I think they're just going to make their employees find other creative ways of using AI without telling them,” he warns.</p><p>Instead, enterprises will want to give employees access to AI tools that help them do their jobs more efficiently, integrate well with existing systems and, ideally, are easy to use.</p><p>As with shadow IT before it, shadow AI will continue to grow if approved alternatives are too limited, too restrictive, or even just too dull.</p> ]]></dc:content>
                                                                                                                                            <link>https://www.itpro.com/technology/artificial-intelligence/in-from-the-shadows-ais-uncontrolled-growth</link>
                                                                            <description>
                            <![CDATA[ AI is now part of our day-to-day language, but the chatbots and generative AI tools spreading through the enterprise are often unauthorised and uncontrolled ]]>
                                                                                                            </description>
                                                                                                                                <guid isPermaLink="false">7wm8gxqYwW5rutFhsXzZgR</guid>
                                                                                                <enclosure url="https://cdn.mos.cms.futurecdn.net/3PVxvzeFqpYHe8FeqsAbUW-1920-80.jpg" type="image/jpeg" length="0"></enclosure>
                                                                        <pubDate>Fri, 20 Feb 2026 11:31:36 +0000</pubDate>                                                                                                                                <updated>Thu, 19 Nov 2026 12:33:34 +0000</updated>
                                                                                                                                            <category><![CDATA[AI Governance]]></category>
                                                    <category><![CDATA[Artificial Intelligence]]></category>
                                                                                                                    <dc:creator><![CDATA[ Stephen Pritchard ]]></dc:creator>                                                                                                        <dc:description><![CDATA[ null ]]></dc:description>
                                                                                                                                <cf:isSponsored>false</cf:isSponsored>
                <cf:hasAffiliateLinks>false</cf:hasAffiliateLinks>
                <cf:isPaid>false</cf:isPaid>
                                                                                                                                                                                    <media:content type="image/jpeg" url="https://cdn.mos.cms.futurecdn.net/3PVxvzeFqpYHe8FeqsAbUW-1920-80.jpg">
                                                            <media:credit><![CDATA[Getty Images]]></media:credit>
                                                                                                                                                                                                                                    <media:description><![CDATA[A graphic of a bar chart with points labelled &#039;AI&#039; on a blue background, to show AI adoption.]]></media:description>                                                            <media:text><![CDATA[A graphic of a bar chart with points labelled &#039;AI&#039; on a blue background, to show AI adoption.]]></media:text>
                                <media:title type="plain"><![CDATA[A graphic of a bar chart with points labelled &#039;AI&#039; on a blue background, to show AI adoption.]]></media:title>
                                                    </media:content>
                                                    <media:thumbnail url="https://cdn.mos.cms.futurecdn.net/3PVxvzeFqpYHe8FeqsAbUW-1920-80.jpg" />
                                                                                                                                                                    <content:encoded >
                            <![CDATA[
                            <article>
                                <p>Artificial intelligence is one of the fastest-growing technologies of all time.</p><p>The United Nations’ trade and development agency, UNCTAD, <a href="https://unctad.org/news/ai-market-projected-hit-48-trillion-2033-emerging-dominant-frontier-technology"><u>predicts</u></a> that the global AI market will be worth$4.8 trillion by 2033, or 25 times its value just three years ago. </p><p><a href="https://www.itpro.com/technology/artificial-intelligence/openai-just-revealed-what-people-really-use-chatgpt-for-and-70-percent-of-queries-have-nothing-to-do-with-work">ChatGPT</a>, which popularised generative AI and <a href="https://www.itpro.com/technology/artificial-intelligence/generative-ai-vs-large-language-models">large language models (LLMs)</a>, reached 100m active users within just two months of its launch in 2022, making it the fastest-growing consumer app ever.</p><p>As ChatGPT’s growth suggests, however, the initial uptake of generative AI has been largely outside corporate IT, with CIOs and enterprises playing catch up ever since.</p><p>Fast-growing, free-to-use large language models and generative AI tools have created a new layer of technology, largely outside the control of corporate IT. This so-called shadow AI has grown, even as enterprise software vendors have added AI functions to their applications.</p><p>While <a href="https://www.itpro.com/technology/artificial-intelligence/gartner-says-40-percent-of-enterprises-will-experience-shadow-ai-breaches-by-2030-educating-staff-is-the-key-to-avoiding-disaster"><u>shadow AI</u></a> has parallels to the shadow IT challenges of the last two decades, uncontrolled use of AI is bringing its own, significant challenges.</p><p>These include breaches of privacy and confidentiality, loss of intellectual property, security issues, and even incorrect and damaging decisions made by shadow AI.</p><h2 id="hiding-in-plain-sight">Hiding in plain sight</h2><p><a href="https://www.gartner.com/en/newsroom/press-releases/2025-11-19-gartner-identifies-critical-genai-blind-spots-that-cios-must-urgently-address0" target="_blank"><u>According to industry analysts Gartner</u></a>, IT leaders in 69% of organizations suspect, or know, that staff are using “unauthorised AI”. The firm also predicts that 40% of businesses will see “security or compliance incidents” resulting from unauthorised AI. </p><p>Security and privacy experts are increasingly worried about the downsides of unauthorised AI, especially as it spreads from simple chatbots to more critical areas such as code development (vibe coding), or <a href="https://www.itpro.com/technology/artificial-intelligence/how-ai-agent-boss-is-reshaping-it-accountability"><u>“agentic” systems</u></a> that operate with little in the way of human oversight.</p><p>Current IT and data security systems are not well-placed to detect and block AI tools experts warn.</p><p>“AI has been available for about three years now, and as such, people are getting used to which platforms they get good results from,” James Gillies, head of cyber security at Logicalis UKI, tells <em>ITPro</em>.</p><p>“It’s become ‘easy’ to get a quick answer to almost any question, but oftentimes the consideration of asking the question in the first place gets forgotten, or worse still, ignored,” he warns.</p><p>This sees employees putting confidential and sensitive information into AI tools, with little or no consideration about where that data goes and how it might be used or reused.</p><p>Employees will continue to use shadow AI, unless CIOs and CISOs can implement strict control measures. Even then, enterprises need to balance the dangers of AI, with the risk of stifling innovation.</p><p>“Unlike traditional shadow IT, which typically involves unauthorised apps or cloud services, shadow AI introduces unique risks because of the way AI tools operate and interact with sensitive data,” warns Findlay Whitelaw, a cybersecurity strategist and researcher at security company Exabeam.</p><p>“Information shared with AI tools may be stored, reused or transferred across jurisdictions often outside organizational visibility or control,” she tells <em>ITPro</em>. “What adds to the complexity of shadow AI is just how widespread its consequences can be.”</p><iframe allow="" height="200px" width="100%" id="" style="" class="position-center" data-lazy-priority="low" data-lazy-src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/" src="https://player.captivate.fm/episode/b6147684-d254-40d8-9409-5f76f5abd749/"></iframe><h2 id="inside-the-machine">Inside the machine</h2><p>For CIOs and those in the enterprise tasked with data security and privacy, there is an urgent need to break down the problem posed by shadow AI. This is not easy.</p><p>“Shadow AI is a bigger concern than shadow IT as it has identity, access and infrastructure impact,” Jon Collins, field CTO at analysts GigaOm, tells ITPro.</p><p>“Shadow IT came with risks: accessing corporate data on unsecured personal devices for example, or using Dropbox and the like for file exchange with data leakage potential, or SaaS systems for project and data management. Each had an operational overhead and was – and remains – a source of pain,” he says.</p><p>“Shadow AI, when confined to the LLM or chatbot world, has similar risks. Additional risks come from agentic approaches, which operate autonomously and at scale. [It’s] no longer a single person causing issues or creating risk, but a single person, potentially inadvertently, launching a flotilla of uncontrolled, autonomous operators into an infrastructure environment which has not been set up to protect against them.</p><p>“Our internal infrastructures have gaping holes of misconfigured systems, poor deprovisioning and access practices, open APIs and mismanaged secrets, which agents can catalogue, navigate and exploit.”</p><p>As Collins describes it, many of the risks from shadow AI are similar to those posed by any unauthorised and uncontrolled technology. For CIOs, this is not a new challenge. Adopting consumer IT in the enterprise has always been a trade off between control and usability, but AI amplifies the problem, both through its complexity and because humans tend to trust machines.</p><h2 id="issues-of-trust">Issues of trust</h2><p>“Shadow AI often exploits what is often described as ‘machine trust’, the assumption that algorithms are inherently accurate and reliable,” suggests Exabeam’s Whitelaw. “Despite this default trust, AI agents are only as good as the data they’re trained on and can carry biases, errors, or vulnerabilities that can be inadvertently, or even intentionally, exploited.”</p><p>Even where individual users trust personal AI tools and perhaps prefer them to corporate technology, they are unlikely to have insights into AI’s algorithms, or details of the data vendors use to train models.</p><p>This, in turn, can lead to unexpected and unwelcome results and the enterprise can be held liable. Controlling shadow AI needs monitoring tools, mapping AI tools to data and business risk, and workforce education.</p><p>“Organizations will have to look into the usage of AI models. To get more visibility into the models they use, they will have to manage usage and make sure they’re not introducing malicious models into their infrastructure,” Ron Minis, senior security researcher at JFrog, tells <em>ITPro</em>.</p><p>However, <a href="https://www.itpro.com/technology/artificial-intelligence/shadow-ai-can-be-a-tool-for-ai-innovation-with-the-right-controls-say-gartner-analysts"><u>banning AI won’t work</u></a>, says Minis. “If companies think they can block or somehow prevent the usage of AI In the company, I think they're just going to make their employees find other creative ways of using AI without telling them,” he warns.</p><p>Instead, enterprises will want to give employees access to AI tools that help them do their jobs more efficiently, integrate well with existing systems and, ideally, are easy to use.</p><p>As with shadow IT before it, shadow AI will continue to grow if approved alternatives are too limited, too restrictive, or even just too dull.</p>
                                                            </article>
                            ]]>
                        </content:encoded>
                                                </item>
            </channel>
</rss>