Security researchers have leveraged Anthropic’s Claude AI to successfully port a pre-authentication remote code execution (RCE) exploit between two models of WAGO programmable logic controllers (PLCs), executing attacker-supplied ARM shellcode on active industrial hardware. This work targets CVE-2021-31886, a critical stack-based buffer overflow in the Nucleus FTP server, which allows unauthenticated access via the FTP USER command over TCP port 21 and carries a CVSS score of 9.8. The research was conducted by Forescout’s Vedere Labs.
How Claude Was Put to Work
The task involved taking an already working exploit for the WAGO 750-852 model and adapting it to another model, the WAGO 750-831 with firmware V01.04.16. Researchers provided Claude with the existing exploit, the firmware binary for the target model, and a live controller to test on. Interactive sessions between a human researcher and Claude Code included access to tools like Ghidra and a terminal connected to the PLC. The work commenced with Claude Sonnet 4.6 and later switched to Claude Opus 4.6 after initial attempts stalled.
On model 750-831, FTP normally zeroed out 256 bytes in an attacker-controlled buffer, preventing shellcode execution. To work around this, Claude devised a modified USER and CWD sequence and omitted the CRLF terminator in one command. That change altered the FTP command processing path such that the buffer wasn’t overwritten in time, enabling execution of the injected payload. Once code execution was established, different payloads were deployed in quick succession: one to send ICMP echo requests to an attacker-controlled host, another to transmit a UDP packet containing the string “PWNED.”
Costs, Risks, and Fallout
The exploit porting process was achieved during an 8-hour-32-minute session and consumed $535.74 in Claude API usage. An attempt to extend the exploit into a command-and-control (C2) implant ended disastrously when writing to flash-mapped memory brick-ed the target PLC. Researchers acknowledged they might have ported the exploit without AI with potentially lower cost, less time, and with the hardware intact.
A real-world risk arises from the fact that neither WAGO nor Siemens have released patches for many of the affected models and firmware versions. CERT@VDE has issued an advisory listing dozens of WAGO PLCs—many running Nucleus V1 RTOS—and warned that updates for those devices are not available. Siemens similarly states that Nucleus NET across all versions will not receive remediation, though later versions of Nucleus ReadyStart since August 2013 address the vulnerability.
During the initial session, Claude also spotted what appears to be a separate vulnerability in FTP’s command extraction loop. While this potential flaw does not yet have a CVE identifier and was excluded from the main exploit work, researchers consider it worthy of further review.
The broader context: Industrial Control Systems (ICS) and OT networks already face threats from chained PLC exploits. In 2021 Forescout exposed 13 hackerable flaws known as NUCLEUS:13. More recently, a joint advisory from U.S. federal agencies warned that threat actors are using AI to generate ICS exploitation tools, lowering the barrier to deploying attacks.
Advisories urge operators of vulnerable WAGO models—750-829, 750-831, 750-852, 750-880, 750-881, 750-882, 750-885, 750-889, 750-331, 750-352 and related devices—to disable or block FTP on port 21, enforce network segmentation, and monitor for anomalous traffic. As of now, many models remain unpatched.
What This Means and What to Watch
This research highlights how AI tools like Claude can accelerate the porting of existing exploits to unpatched devices and even expose new vulnerabilities—turning proof-of-concept into practical threats in operational technology environments. It signals a shift: attackers may soon need less specialized firmware knowledge when AI can adapt known exploits across different hardware. IT and OT teams must treat threat modeling for AI-assisted exploitation as an urgent priority. Moving forward, watch for more reports where attackers use AI to scale exploit generation, plus how device manufacturers respond with firmware updates or redesigns to mitigate AI-accelerated threats.
