What's Happening?
Researchers at Forescout’s Vedere Labs successfully used Anthropic’s Claude AI to port a remote code execution (RCE) exploit from one WAGO programmable logic controller (PLC) to another. The experiment aimed to adapt a previously developed exploit for
the WAGO 750-852 PLC (CVE-2021-31886) to a related model, the WAGO 750-831, and to explore the AI's ability to develop a full command-and-control implant. Claude Code was provided with access to a terminal, reference files, the reverse-engineering tool Ghidra, and the physical target device. The AI initially confirmed the vulnerability by generating a payload that crashed the PLC. However, achieving controlled code execution proved more challenging, requiring significant researcher oversight and redirection. A breakthrough occurred when the team switched from Claude Sonnet 4.6 to Claude Opus 4.6 and instructed the AI to request help when uncertain about firmware details. This led to the AI identifying and overcoming an issue where injected code was being erased, subsequently producing two working payloads within 12 minutes. The experiment, however, was time-consuming, taking hours, and incurred over $500 in API costs for the RCE development phase alone. A follow-up attempt to build a command-and-control implant resulted in the AI bricking the device by writing to its flash memory.
Why It's Important?
This experiment highlights the dual nature of AI in cybersecurity: its potential as a powerful tool for exploit development and its current limitations requiring significant human intervention and cost. The ability of AI to analyze firmware, identify vulnerabilities, and generate functional exploits, even with guidance, suggests a future where such capabilities could accelerate the development of cyber weapons. This has significant implications for the security of critical infrastructure, particularly industrial control systems like PLCs, which are increasingly targeted by cyberattacks. The high cost and time involved, coupled with the AI's capacity to 'brick' a device, underscore that while AI can assist, it is not yet an autonomous solution for complex cybersecurity tasks. This research provides valuable insights into the evolving threat landscape, indicating that defenders must anticipate AI-assisted attacks and develop corresponding AI-driven defense mechanisms. The findings also suggest that the marginal cost of scaling such work with AI could decrease over time, making it a more accessible tool for both malicious actors and security researchers.
What's Next?
The findings suggest that the role of human researchers in guiding and correcting AI will remain crucial for the foreseeable future in complex tasks like exploit development. However, as AI models improve and become more sophisticated, the level of human intervention required is likely to decrease, potentially making AI-assisted exploit generation more efficient and cost-effective. This could lead to a proliferation of AI-driven cyberattacks, necessitating a corresponding advancement in AI-powered defensive strategies. Organizations responsible for critical infrastructure, such as water systems and industrial control systems, will need to invest more heavily in robust cybersecurity measures and potentially integrate AI into their defense strategies to counter these evolving threats. Further research will likely focus on improving AI's ability to handle complex, nuanced tasks in cybersecurity, while also developing safeguards to prevent unintended consequences like device bricking.
Beyond the Headlines
Beyond the immediate technical implications, this experiment touches upon the ethical considerations of AI in offensive cybersecurity. The ease with which AI can be directed to identify and exploit vulnerabilities raises questions about responsible AI development and deployment. The potential for AI to automate and scale cyberattacks could dramatically alter the balance between attackers and defenders, potentially leading to more frequent and sophisticated breaches. This necessitates a broader discussion on international norms and regulations for AI in cyber warfare. Furthermore, the experiment highlights the importance of 'explainable AI' in critical applications; understanding why an AI makes certain decisions, especially when dealing with sensitive systems, is crucial for debugging, auditing, and ensuring accountability. The incident of the AI bricking the device also serves as a stark reminder of the unpredictable nature of advanced AI systems and the need for fail-safes and human oversight in high-stakes environments.
