AIは自律的にハッキングできるのか?研究者が「Human in the Loop」の不可欠性を発見
セキュリティ研究者のJames Kettle氏は、AIは強力であるものの、人間の介入なしに新しいハッキング戦略を自律的に開発することはできないことを発見した。
in research presented at black hat, james kettle tested whether agentic ai could autonomously conceptualize new hacking strategies from scratch. he concluded that ai is minimally capable of independent invention but is extremely powerful when guided by human insight. this human-ai collaboration led to the discovery of shared-parser confusion, a new vulnerability in web servers using shared code. the research highlights that ai security tools currently function best as force multipliers for human experts rather than independent agents.