Anthropic reported that users in Houthi-controlled northern Yemen utilized its Claude AI model to attempt the development of advanced missile systems, including hypersonic and guided rocket technology.

Key Points

• Anthropic blocked accounts in northern Yemen after discovering users were leveraging Claude Code to develop guidance, navigation, and control software for missiles.
• The users attempted to build a multi-variant missile capable of hypersonic speeds and a warhead utilizing mobile phone hardware for mid-course maneuvering.
• While the actors failed to field an operational device, they successfully conducted a test of a guided rocket and returned to the chatbot to troubleshoot failures.
• The findings were detailed in Anthropic’s third report on AI misuse, which also covered state-sponsored propaganda and research into biological weapons.
• Weapons analysts noted that while the Houthis lack the industrial capacity to build hypersonic missiles, the attempt signals a push for greater military self-reliance.

Why it Matters

This incident highlights the growing security risks associated with the accessibility of advanced generative AI tools in conflict zones. It underscores the urgent need for AI developers to implement robust safeguards to prevent non-state actors from accelerating the development of sophisticated weaponry.