How Yemeni rebels used Anthropic's AI software to design and build guided weapons

In Yemen, Houthi rebels have been utilizing Anthropic's AI software to design and develop guided weapons, according to a detailed report released by the company. This ambitious weapons development effort took place between December 2025 and August 2026 and involved three missile projects simultaneously. The Houthis, who control the capital Sanaa and northern Yemen, expanded their reach by seizing the strategic port city of Mokha on the western coast, further extending their control along the Red Sea and near the Bab Al Mandeb shipping route.

The report, titled "Detecting and countering misuse of AI," revealed astonishing detail about the group's work, showcasing the scale and sophistication of their ambitions. One project aimed to transform a commercially available rocket into a guided missile using off-the-shelf technology and a homing system. Another project sought to create a multi-stage ballistic missile capable of traveling over 2,000 kilometers, while the most striking initiative, dubbed the "R2000" program, aimed to develop a family of missiles including a hypersonic glide vehicle, a technology typically associated with advanced military powers.

The group reportedly used Anthropic's AI software, specifically the Claude Code, to write the guidance, navigation, and control software for these missiles. They organized their work similarly to a team, with different instances of Claude handling distinct tasks such as coding, research, and reviewing the work of others. Despite the AI safeguards in place, the group managed to work around them by concealing their activities and breaking their work into separate sessions, thus avoiding detection.

However, they were unable to create a fully functional weapon, as far as Anthropic could determine. They did, though, test-fire a guided rocket, which failed within hours. The group circumvented the ban by developing an offline simulation toolkit that operates without Claude or commercial tools like MATLAB, ensuring some of their capabilities persisted even after the ban.

In response to the discovery, Anthropic banned the accounts involved and shared threat information with government and industry partners. This development comes amid growing concerns over the increasing power and potential risks of AI models, following several high-profile incidents over the past decade.