Anthropic says Iran used Claude to target US Navy ships

An operation linked to Iran fed ship photos, transponder signals, and satellite images into Claude, Anthropic's chatbot. The goal was finding exactly where U.S. Navy ships sat in the Persian Gulf. Anthropic disclosed this Thursday, in its third report on how people misuse its AI. A separate case in the same report: a cell in Yemen used Claude to help design missile guidance systems.

Anthropic didn't publicly name the group. The company says it detected and disrupted the operation. That's Anthropic's account, not something outside investigators have confirmed. This is Anthropic's third public misuse report. An earlier one, from August 2025, covered a North Korean worker fraud scheme. This new report also details a bioweapons case Anthropic separately disclosed this week.

Keep reading — what to watch

Will other AI companies start publishing reports like this? Right now, mostly Anthropic does this kind of detailed public disclosure. OpenAI faced its own rogue-agent misuse claims this month, over a hacking incident, not a military one. The real question is which AI tool a hostile government tries next, and whether the public ever hears about it.