Sana'a: A cell of threat actors based in northern Yemen has covertly utilized the Claude artificial intelligence system by Anthropic to aid in the development of guided rockets and other sophisticated weaponry. According to a report from the AI company, the group employed Claude Code to create software for guiding and stabilizing flying vehicles, while deliberately hiding the true purpose of their work.
According to TRTworld.com, although the Anthropic report does not explicitly mention the Houthis, the Iran-aligned group controls northern Yemen. The actors were reportedly engaged in three weapons programs: a guided rocket using a phone-class flight computer with homing guidance, a multi-stage ballistic missile aimed at a range exceeding 2,000 kilometers, and a missile program, the R2000 set, featuring a hypersonic glide vehicle variant.
The report highlights that the group replaced human software engineers with Claude Code to develop software for guidance, navigation, and control. Claude was instrumental in various development stages, including software writing, research, flight control tuning, software builds, and flight simulations. The actors divided their tasks among multiple Claude sessions, with specific instances dedicated to coding, research, and quality review.
Anthropic's safeguards managed to block many of the group's requests, yet the actors found methods to bypass certain restrictions. They concealed their true objectives and distributed their work to prevent any single conversation from revealing the entire scope of the weapons program.
The group's efforts culminated in a live test of a guided rocket in Yemen, which reportedly failed. Within hours, the actors consulted Claude to investigate the failure. Anthropic found no evidence that the group deployed an operational weapon but identified ongoing efforts to simulate and develop weapons beyond the rocket program. Claude was used in ballistic missile simulations, trajectory modeling, and flight-control system optimization.
Furthermore, the actors created an offline simulation toolkit independent of Claude or other engineering software, allowing for continued engineering efforts without relying on Anthropic's AI system. Following these discoveries, Anthropic banned accounts linked to the actors and shared their findings with public and private partners. The case highlights the challenges of preventing AI-assisted weapons development when users intentionally obscure their objectives and distribute complex projects across multiple interactions.
The report emerges amid increasing tensions, as Houthis intensify attacks on Saudi Arabia's oil facilities, prompting airstrikes from Riyadh. The Iran-backed group has also seized control of Yemen's strategic coastal city of Mocha, as conflicts with the internationally recognized Yemeni government persist.