ANTHROPIC has said that Claude users in northern Yemen, an area controlled by Iran-backed Houthi rebels, attempted to use the AI model to help develop more sophisticated weaponry. While the firm identified and blocked these users, it stated they did not manage to field an operational device, but did conduct a failed test of a guided rocket.
The company says the users sought guidance from Claude by returning to the chatbot to understand why the test failed, and that the evidence included an offline simulation toolkit they had prepared that did not rely on Claude or similar platforms. The finding comes from a December-to-August assessment and points to efforts to pursue three different weapon programmes, including a multi-variant missile capable of gliding at hypersonic speeds and a warhead that uses mobile phone hardware to manoeuvre mid-course.
Anthropic’s report also notes the broader context: the Houthis, who already wield drones and missiles and have been pressing for greater control of territory and disruption of regional shipping, may be exploring more capable weapons to reduce reliance on Iranian-supplied components. Trevor Ball, a weapons analyst, cautioned that while the Houthis might be exploring hypersonics, their production and technical capabilities are not comparable to advanced systems in testing today.
The company disclosed it shared its findings with private and public partners. The piece highlights ongoing concerns about AI’s role in warfare and the potential for actors to experiment with automated toolkits in pursuit of more capable munitions.