Anthropic says Houthi‑held Yemen tried to use Claude to design advanced weapons

US AI firm Anthropic has disclosed that its conversational model Claude was nearly exploited by users in territory controlled by Yemen’s Houthi movement to assist in developing advanced weapons, including missiles. The Houthis, an Iran‑aligned armed group already under international sanctions for attacks on shipping in the Red Sea, allegedly submitted prompts seeking detailed guidance on missile propulsion and guidance systems.

Anthropic says its internal security systems flagged the activity, leading the company to shut down the accounts and notify relevant authorities. The firm has long promoted strict “guardrails” that block dangerous weapons‑related queries, but this incident highlights how hard it is to separate legitimate engineering questions from covert attempts to obtain high‑end military know‑how through generative AI.

The case is a concrete example of how modern AI tools can be pulled into the orbit of state and non‑state actors seeking more sophisticated weaponry. It is likely to fuel calls for tighter international controls on AI access, from export rules and geofencing to mandatory monitoring and third‑party audits, as governments grapple with the dual‑use nature of frontier models and the risk that they could accelerate real‑world conflict.