Security researchers warn of AI-powered PLC attacks in wake of Siemens advisories

Security researchers have successfully used AI to port a remote code execution (RCE) exploit between two programmable logic controllers (PLCs).

While the experiment still required significant human expertise to negotiate dead ends and false leads, it showed how AI could make it easier to carry out attacks against embedded and industrial systems.

The team at Forescout’s Vedere Labs used AI to port an RCE exploit between two WAGO PLC models in an exploit that took eight hours and 32 minutes and consumed just $535.74 in API tokens.

Once code execution was achieved, AI produced multiple working network payloads within minutes, suggesting that post-exploitation could become increasingly automated as models improve.

The exploit targeted CVE-2021-31886, a pre-authentication buffer overflow in the Nucleus FTP server that allowed arbitrary ARM shellcode to execute on the live PLC without credentials.

Researchers had already shown that RCE exploits on PLCs can enable Deep Lateral Movement and granular control over safety logic – and reckoned that as AI-assisted exploit development improves, these techniques could become more accessible.

How AI was used to crack PLCs

The AI-assisted exploit development process involved two steps, according to Forescout. First and foremost, this included confirming the vulnerability and writing the payload.

Each step consisted of interactive sessions between a researcher and Claude Code, which had access to a terminal, the reference files, analysis tools including Ghidra, and the live target PLC.

Claude could use those tools directly, generate and test code, and ask the researcher for additional input when needed.

Researchers noted that Claude decided to confirm the presence of the vulnerability by both probing the live target and carrying out static code analysis.

It didn't work the first time, however, with one session failing to trace the vulnerable function correctly and producing an invalid exploit.

According to researchers, the analysis provided useful context for the next session, which used binary searches to map the relevant functions correctly – albeit with researcher input to steer the analysis away from dead ends and provide additional disassembly context where necessary.

Writing a working RCE exploit took much longer, with initial attempts failing. Claude also wasted a lot of time testing incorrect hypotheses, decompiling unrelated code, and pursuing false leads.

The researchers had to change to Claude Opus 4.6 with 1M context, and add the prompt: “Ask for my help with disassembly if you are not certain about a firmware detail”, and instruct the AI to the sink of the vulnerability – the point where the attacker-supplied username is copied into memory.

Claude could then reason about the function call chain leading to the sink and understand that it needed more context about how FTP packets are treated throughout this call chain.

Operational technology in the crosshairs

Organizations shouldn't dismiss operational technology (OT) vulnerabilities because they seem hard to exploit, researchers warned.

"AI has already lowered the barrier to vulnerability research and exploit development in higher-level software. This experiment suggests that the same progression is beginning to reach low-level embedded systems, although substantial barriers remain," the report noted..

"As models become more capable and independent, the cost and expertise required to adapt exploits across related embedded targets could fall substantially."

The advice on mitigation is pretty standard: reduce unnecessary OT device exposure, monitor OT environments for early signs of exploitation, exercise incident response against AI-assisted OT attack paths and use AI defensively, but validate its outputs.

Programmable logic controllers are rapidly emerging as a major risk to critical infrastructure. The US National Security Agency (NSA) issued a warning last month over an active threat against Siemens PLCs.

Similarly, in recent months, a series of attacks on US water supply infrastructure targeted internet-exposed PLCs, with the attackers remotely changing IP addresses and turning on and setting passwords.

Several water firms were left unable to view connected equipment, and in some cases it was shut down.

FOLLOW US ON SOCIAL MEDIA

Follow ITPro on Google News and add us as a preferred source to keep tabs on all our latest news, analysis, views, and reviews.

You can also follow ITPro on LinkedIn, X, Facebook, and BlueSky.