The Most Dangerous AI Hacking Techniques Still Have Humans in the Loop

James Kettle presented research at Black Hat showing that agentic AI, when guided by a human, can accelerate vulnerability research and helped identify a novel potential web vulnerability class dubbed "Shared-Parser Confusion." The work demonstrates AI's growing utility as a research partner but also its limitations: the finding was a hypothesis confirmed by the researcher, was rare, not proven exploitable in the tested target, and there is no evidence of active exploitation.