AI Ports PLC Exploit in Hours, Costs Researchers $500 in API Fees

Researchers at Forescout's Vedere Labs successfully used Anthropic's Claude to port a remote code execution (RCE) exploit between two WAGO programmable logic controllers, but the experiment revealed that AI-assisted exploit development remains expensive, slow, and requires heavy human oversight. The starting point was a known exploit for the WAGO 750-852 PLC based on CVE-2021-31886, a pre-authentication buffer overflow in the Nucleus FTP server that allows unauthenticated attackers to execute arbitrary ARM shellcode. The team then tasked Claude with adapting the exploit to a related model, the WAGO 750-831, and subsequently developing a command-and-control implant on the compromised device.

To conduct the experiment, researchers equipped Claude Code with terminal access, reference files, the Ghidra reverse-engineering framework, and a physical target PLC. The AI confirmed the vulnerability through live probing and static firmware analysis before generating a payload that crashed the device. Achieving controlled code execution proved far more difficult, with early attempts leading the model down incorrect technical paths until the team switched from Claude Sonnet 4.6 to Claude Opus 4.6 and instructed it to request clarification on uncertain firmware details. The breakthrough came when Claude identified why injected shellcode was being erased before execution and adjusted the payload placement accordingly, producing two working exploit variants within 12 minutes. A subsequent attempt to develop a full C2 implant went wrong when a test payload wrote to flash memory and permanently bricked the PLC.

The financial and time costs were substantial: the final stage of RCE development alone consumed over $500 in Anthropic API charges across an eight-hour session. Forescout acknowledged that a skilled human researcher could have completed the initial exploit port faster, cheaper, and without sacrificing the hardware, but noted that the trajectory of these costs is worth monitoring as models improve. For defenders tracking ICS threats, the research underscores how quickly AI-assisted exploitation can move from proof-of-concept to operational capability. Security teams managing industrial environments can use tools like a port scanner to audit exposed PLC management interfaces, a SSL/TLS checker to validate encrypted management channels, and a WHOIS lookup to vet third-party vendors with remote access to operational networks.