Houthi rebels used Anthropic's Claude AI to design missiles

Houthis attempted AI-assisted weapons development

Anthropic has disclosed that users in Houthi-controlled northern Yemen attempted to use its Claude AI chatbot to develop advanced missile systems, according to a report released Thursday by the company.

The blocked accounts pursued three weapons programs, including a multi-variant missile designed to glide at hypersonic speeds and a warhead using mobile phone hardware for mid-course maneuvering. The users employed Claude Code as a substitute for human software engineers to develop guidance, navigation, and control software.

While the accounts did not succeed in fielding an operational device, Anthropic said the users conducted a failed test of a guided rocket — then returned to Claude to troubleshoot why it failed. The company blocked the accounts after identifying the activity.

Hazam al-Assad, a member of the Houthis' political bureau, dismissed the report as "unreasonable and illogical," stating that Houthi forces possess "modern, diverse and developed production capabilities" accumulated during Yemen's 12-year civil war.

Technical capabilities and limitations

Weapons analyst Trevor Ball of Armament Research Services expressed skepticism about the Houthis' ability to actually produce hypersonic missiles, noting that even U.S. hypersonic programs remain in testing phases. He suggested the group may be attempting to reduce dependence on Iranian weapons shipments by developing indigenous capabilities.

Before the ban, Anthropic found evidence that the users had built an offline simulation toolkit independent of Claude or other computing platforms, suggesting some technical progress despite the failed test.

The Houthis already possess Iranian-made anti-ship missiles with mid-course guidance adjustment capabilities. Iran denies arming the group in violation of U.N. embargoes, though Iranian weapons have been documented on Yemeni battlefields and intercepted in shipments.

Broader strategic context

The disclosure comes as Houthi forces have made significant territorial gains along the Red Sea coast and Bab al-Mandeb Strait since Wednesday, threatening key shipping routes. The group has conducted waves of attacks on Saudi oil facilities and Red Sea tankers, compounding Iran's disruption of shipping through the Strait of Hormuz.

Adam Baron, a Yemen researcher at the New America think tank, cautioned against underestimating Houthi technical sophistication. "There's a tendency to see the Houthis as this group of barefoot tribal fighters, and that's just not true," he said, pointing to their "incredible — and deepening — amount of institutional tech savvy."

Why it matters

The incident illustrates how frontier AI models are being tested for military applications by non-state actors with limited resources. While the Houthis' attempt failed and experts doubt their capacity to build truly advanced systems, the case demonstrates both the accessibility of AI tools for weapons research and the challenge facing AI companies in preventing misuse. As AI capabilities expand, the gap between asking an AI how to build a weapon and actually producing one remains significant — but the barrier to entry for initial research and development has clearly lowered.

The findings appeared in Anthropic's third misuse report since March 2025, covering activities from December to August. The company said it shared its findings with private and public partners.

These details were first reported by the Associated Press.