FORESCOUT'S Vedere Labs leveraged Anthropic's Claude to successfully adapt a remote code execution (RCE) exploit from the WAGO 750-852 PLC to the WAGO 750-831 model. The experiment followed previous PLC attacks in the water sector and involved extensive oversight and costly use of APIs (over $500). The process required significant teamwork between researchers and the AI, with initial struggles but eventual rapid iteration once breakthroughs were made.
However, a follow-up for developing a command-and-control implant resulted in bricking the device. This highlights the AI's potential to reduce the marginal cost of exploit porting, although it's currently more efficient for human researchers.