Cybersecurity Researchers Demonstrate AI-Powered PLC Exploit Porting in Hours and Dollars
In a stunning demonstration of the accelerating pace of artificial intelligence (AI) in cybersecurity, researchers at Forescout’s Vedere Labs have successfully ported a working remote code execution exploit from one programmable logic controller (PLC) to another using Anthropic’s Claude AI. The experiment, which was conducted in response to recent attacks targeting PLCs in the water sector, highlights the rapidly evolving threat landscape and the growing importance of AI-powered cybersecurity tools.
The starting point for the experiment was a previously developed exploit for the WAGO 750-852 PLC, based on CVE-2021-31886, a pre-authentication buffer overflow vulnerability in the Nucleus FTP server that allows an unauthenticated attacker to execute arbitrary ARM shellcode. The researchers aimed to adapt this exploit to a related but distinct model, the WAGO 750-831, and use Claude AI to push the result further into a full command-and-control implant.
Using Claude Code, the researchers gave the AI access to a terminal, reference files, the reverse-engineering tool Ghidra, and the physical target device. The AI confirmed the vulnerability through a mix of live probing and static firmware analysis before generating a payload that crashed the PLC. However, turning this into controlled code execution took far longer, with early attempts sending the AI chasing incorrect leads.
The breakthrough came when the team switched to Claude Opus 4.6 and instructed the model to ask for help whenever it was uncertain about a firmware detail. The AI then figured out why the injected code kept getting erased before it could run, adjusting its approach to keep the payload intact. Once this obstacle was cleared, Claude produced two separate working payloads within 12 minutes.
The experiment highlights the rapidly evolving threat landscape and the growing importance of AI-powered cybersecurity tools. While the cost of scaling such work is still high, with the final stage of RCE development alone consuming over $500 in API usage across a session lasting more than eight hours, the potential for AI to reduce the marginal cost of doing so across many related targets at once is significant.
The experiment also serves as a reminder that the use of AI-powered tools can have unintended consequences. In a follow-up attempt to build on the RCE, Claude tested increasingly complex payloads, and one of those payloads wrote to a region mapped to the PLC’s flash memory, permanently bricking the device.
As cybersecurity professionals, it is essential to stay ahead of these evolving threats and understand how AI-powered tools can be used for both good and ill. By exploring the capabilities and limitations of such tools, we can better prepare ourselves for the challenges that lie ahead and develop effective strategies to mitigate the risks associated with AI-powered cyberattacks.
Ultimately, this experiment serves as a stark reminder of the accelerating pace of cybersecurity threats and the need for robust, AI-powered defenses to stay ahead of the curve. As researchers continue to push the boundaries of what is possible with AI in cybersecurity, we must remain vigilant and proactive in our approach to protecting ourselves from these emerging threats.
Source: SecurityWeek — 2026-09-01