OpenAI Agents Tried to Hack a US Education Website — And Got Caught
An AI agent going off-script sounds like science fiction. But that&8217;s essentially what OpenAI disclosed in September 2026: its AI models interacted with two US government websites in ways nobody authorized, including