UK releases draft code of practice for AI security

Open source generative AI concept image showing a digitized human brain elevated over a GPU and circuit board.
(Image credit: Getty Images)

The UK government has released two codes of practice to enhance cyber security in AI and software, aimed at making products resilient against tampering, hacking, and sabotage.

The aim of the guidelines, unveiled during the CYBERUK conference, is to improve confidence in the use of AI models across industries while helping businesses innovate.

"We have always been clear that to harness the enormous potential of the digital economy, we need to foster a safe environment for it to grow and develop. This is precisely what we are doing with these new measures, which will help make AI models resilient from the design phase," said technology minister Saqib Bhatti.

"Today’s report shows not only are we making our economy more resilient to attacks, but also bringing prosperity and opportunities to UK citizens up and down the country."

The new codes of practice come as a report shows that the cyber security sector has experienced a 13% growth on last year, and is now worth almost £12 billion - on par with sectors such as the automotive industry.

The codes are based on guidelines from the National Cyber Security Centre (NCSC), published in November last year.

They include measures such as minimizing and anonymizing data use, establishing robust data governance policies, conducting regular audits and impact assessments, securing data environments, and keeping staff up to date with current security protocols.

The government said it hopes the codes will eventually form the basis of a future international standard.

"Plans for it to form the basis of a global standard are crucial, given the central role international standards already play in addressing AI safety challenges through global consensus," said Rosamund Powell, research associate at The Alan Turing Institute.

"Research highlights the need for inclusive and diverse working groups, accompanied by incentives and upskilling for those who need them, to ensure the success of global standards like this."

Kevin Curran, professor of cyber security and senior member of the Institute of Electrical and Electronics Engineers (IEEE), welcomed the move as a positive step toward creating a safer environment for developers and businesses alike.

“Understanding how generative AI systems arrive at their outputs can be difficult,” he said.

“This lack of transparency means it can be hard to identify and address potential biases or security risks. Generative AI systems are particularly vulnerable to data poisoning and model theft.

"If companies cannot explain how their GenAI systems work or how they have reached their conclusions, it can raise concerns about accountability and make it difficult to identify and address other potential risks."

During his speech, Bhatti also announced new initiatives on how the government and regulators will professionalize the cyber security sector, such as incorporating cyber roles into government recruitment and HR policies.

The government is also launching a consultation on scaling up the impact of the CyberFirst scheme, seeking views on its future direction, the creation of a new alternatively-led organization to take over CyberFirst delivery and achieve scale, and the role of government in supporting the delivery of the scheme in the longer term.

Emma Woollacott

Emma Woollacott is a freelance journalist writing for publications including the BBC, Private Eye, Forbes, Raconteur and specialist technology titles.