Frontier AI can be used to exploit flaws in programmable logic controllers, which are critical devices that were targeted in a recent hacking campaign against U.S. water utilities, energy companies and factory operations, according to a report Tuesday from Forescout Research - Vedere Labs.
Researchers used Claude to port a remote code execution vulnerability from a PLC model made by WAGO Corp. to another device using the same model. The vulnerability, tracked as CVE-2021-31886, which is a pre-authentication overflow in the Nucleus FTP server.
The simulated attack — conducted in a test environment — demonstrated that AI could be used to lower the barrier for an attacker. Still, researchers noted that significant hurdles remain before such an attack could be replicated without human guidance.
“The AI helped to confirm the existence of the vulnerability on the other PLC model and to construct an exploit for it,” Daniel dos Santos, VP of research at Vedere Labs, told Cybersecurity Dive. “The AI did not manage to construct the exploit entirely autonomously, as it needed the researcher’s help to focus on what to exploit and how.”
Forescout previously developed a working remote code execution exploit for the WAGO 750-852, documented in a prior report. In the current test, researchers tried to port the exploit to a similar model, WAGO 750-831.
Researchers first tried Claude Sonnet 4.6. That attempt ran into difficulties, including a failure to trace the vulnerable function in the PLC, which resulted in an invalid exploit, according to the report. Claude eventually was able to crash 750-831, but struggled to write a working RCE code.
When researchers switched to Claude Opus 4.6, they were able to figure out why the shellcode was not surviving long enough to execute. In total, eight hours were required to execute working code. Researchers helped navigate around a number of false leads and dead ends.
After Claude figured out problems with buffer preservation, it was able to create two functional RCE payloads in 12 minutes. Despite that success, Claude managed to brick the PLC when it attempted to improve on the exploit, according to researchers.
The report comes just weeks after suspected Iran-linked hackers targeted drinking and wastewater utilities in 12 U.S. states. Authorities said the attackers were able to change passwords and temporarily render operators unable to monitor their systems. This forced some operators to briefly suspend water service or deal with sewage floods.
On Monday, the White House launched Project Watershed 250, a program designed to help water utilities harden their OT environments with the help of private-sector security experts.
