Sanaa: A group of threat actors based in northern Yemen has covertly harnessed the Claude artificial intelligence system developed by Anthropic to aid in the creation of guided rockets and other sophisticated weaponry, detailed in a report by the AI company.
According to TRTworld.com, while the report from Anthropic did not specifically identify the Houthis, the region in question is under the control of the Iran-backed group. The actors leveraged Claude Code to develop software for guiding and stabilizing aerial vehicles while attempting to obscure their true objectives from the AI company. This was disclosed in Anthropic's September 2026 report on identifying and countering the misuse of AI technologies.
Anthropic revealed that the group was engaged in three separate weapons programs: the development of a guided rocket using a phone-class flight computer with final-phase homing guidance, a multi-stage ballistic missile with a range exceeding 2,000 kilometers, and a missile program known as the R2000 set, featuring a hypersonic glide vehicle variant. The actors employed Claude Code in lieu of human software engineers for the guidance, navigation, and control software essential for these programs.
The AI system was integral to numerous facets of the weapons' development process, including software writing, research conduction, flight control tuning, software builds, and flight simulations. The Houthis managed multiple concurrent Claude sessions, assigning distinct tasks to separate AI instances-one for coding, another for research, and a third for reviewing the outcomes produced by the initial instance.
Anthropic noted that although its safeguards thwarted many of the group's requests, they managed to circumvent some restrictions. The group's strategy involved concealing their objectives and distributing their work across multiple sessions to prevent any single conversation from disclosing the full extent of their weapon development efforts.
Despite these efforts, the group's live test of a guided rocket in Yemen seemingly failed, as reported by Anthropic. Within hours, the actors returned to Claude to analyze the failure, yet Anthropic found no evidence of a successfully operational weapon being fielded. Nonetheless, the company uncovered ongoing efforts to develop and simulate additional weapons beyond the original rocket program.
The actors also devised an offline simulation toolkit that functioned independently of Claude and other engineering software like MATLAB, culminating in a standalone program that persisted without the AI system. This advancement enabled the group to continue its engineering endeavors without relying on Anthropics AI technology.
In response, Anthropic banned the accounts associated with these actors and disseminated information regarding their activities to partners in the public and private sectors. The case highlights the challenges in preventing AI-assisted weapons development when users intentionally obscure their objectives and divide complex projects across multiple interactions.
This revelation comes amid escalating tensions as the Houthis increase attacks on Saudi Arabia's oil infrastructure, prompting retaliatory airstrikes from Riyadh. The Iran-backed group has also seized control of Yemen's strategic coastal city of Mocha, intensifying conflicts with Yemen's internationally recognized government.