Anthropic says AI helped Houthis develop missiles with guidance systems

The actors used Claude Code to integrate autopilot software, write control algorithms, and run firmware builds. After a failed test-fire of the guided rocket, they returned to the AI within hours to diagnose issues. Anthropic blocked many requests through safety controls, but the actors evaded them by splitting tasks across multiple sessions and hiding their true intent. The report notes that once code and workflows were transferred to local tools, platform bans became ineffective. Anthropic banned linked accounts and shared findings with partners but did not confirm the actors’ affiliation with the Houthi movement, though external reports suggest a possible link.