Experiment: Porting a PLC Exploit With AI Takes Hours and Hundreds of Dollars

Researchers at Forescout's Vedere Labs used Anthropic's Claude AI to adapt a known RCE exploit (CVE-2021-31886) from the WAGO 750-852 PLC to the related 750-831 model. The exploit abuses a pre-authentication buffer overflow in the Nucleus FTP server to execute arbitrary ARM shellcode. The AI was given access to tools including a terminal, Ghidra, reference files, and the physical device. It confirmed the vulnerability through live probing and static analysis, then generated payloads that caused crashes and eventually controlled code execution. The process required several hours of researcher guidance and over $500 in API usage. Attempts to extend the exploit into a command-and-control implant led to a payload that permanently bricked the PLC by writing to flash memory. The experiment demonstrates AI's potential to accelerate exploit porting but also the need for expert oversight and the risks of unintended device damage.

The experiment confirms that the WAGO 750-831 PLC shares the same vulnerability (CVE-2021-31886) as the 750-852 model, allowing unauthenticated remote code execution via a buffer overflow in the Nucleus FTP server. Successful exploitation could enable attackers to execute arbitrary ARM shellcode on the device. However, the experiment also shows that developing reliable payloads is complex and error-prone, with risks such as device bricking. While no active exploitation in the wild is reported, the demonstrated vulnerability could be leveraged in targeted attacks against these PLCs, potentially impacting industrial control systems.

Patch status is not confirmed in the provided data. Users should consult WAGO vendor advisories for updates regarding CVE-2021-31886 and related vulnerabilities. Until official patches are applied, organizations should consider network segmentation and restrict access to PLC management interfaces to trusted personnel and systems. Given the experimental nature of the AI-assisted exploit porting, no immediate new mitigation beyond existing CVE-2021-31886 guidance is indicated.