Sana'a: A cell of threat actors based in northern Yemen has allegedly utilized the Claude artificial intelligence system by Anthropic to secretly advance their weapons program, focusing on developing guided rockets and other sophisticated armaments.
According to TRTworld.com, while the Anthropic report does not directly mention the Houthis, the northern region of Yemen is predominantly under the Iran-aligned group's control. The report revealed that the group employed the Claude AI system to craft software intended for the guidance and stabilization of flying vehicles. Despite attempts to hide their true objectives from Anthropic, the company outlined these activities in its September 2026 report on detecting and countering AI misuse.
Anthropic's findings highlighted that the actors were engaged in three separate weapons initiatives: the creation of a guided rocket with a phone-class flight computer and homing guidance, a multi-stage ballistic missile with an intended range exceeding 2,000 kilometers, and a missile program called the R2000 set, which includes a hypersonic glide vehicle variant. The use of Claude Code replaced human software engineers for developing guidance, navigation, and control software.
Claude was integral to various stages of the development process, including writing software, conducting research, adjusting flight controls, managing software builds, and executing flight simulations. The Houthis reportedly ran multiple Claude sessions at once, partitioning tasks across different AI instances for coding, research, and review purposes.
Anthropic noted that although its safeguards blocked several requests from the group, some restrictions were circumvented. The actors endeavored to keep their objectives hidden and distributed their work over multiple sessions, ensuring no single conversation exposed the entire scope of their weapons program.
The group reportedly conducted a live test of a guided rocket in Yemen, which appeared to fail. Following the test, the actors promptly returned to Claude to analyze the failure. Anthropic has not found evidence of any successful deployment of operational weapons, but there is proof of ongoing efforts to develop and simulate weapons beyond the rocket program.
Anthropic disclosed that Claude was also used for ballistic missile simulations and other weaponry, including modeling trajectories and optimizing flight-control systems. The actors further developed an offline simulation toolkit independent of Claude or engineering software like MATLAB. This toolkit evolved into a standalone program, enabling the group to progress their engineering efforts without relying on Anthropic's AI system.
Subsequently, Anthropic banned accounts linked to the actors and shared details of their activities with public and private sector partners. The company emphasized the challenge of curtailing AI-assisted weapons development when users deliberately obscure their goals and divide intricate projects across multiple interactions.
These revelations emerge amidst heightened tensions, with Houthis increasing attacks on Saudi Arabia's oil facilities and Riyadh responding with airstrikes. The Iran-backed group has also seized control of Yemen's strategic coastal city of Mocha, as conflict persists with Yemen's internationally recognized government.
