A “cell of threat actors” based in northern Yemen secretly used Anthropic’s Claude artificial intelligence system to help develop a guided rocket and other advanced weapons, according to a report by the AI company.
While the Anthropic report did not explicitly name the Houthis, northern Yemen is under the control of the Iran-aligned group.
The group used Claude Code to develop software responsible for guiding and stabilising flying vehicles, while attempting to conceal the purpose of its work from Anthropic, the company said in a September 2026 report on detecting and countering AI misuse.
Anthropic said the actors were working on three weapons programmes: a guided rocket using “a phone-class flight computer” and final-phase homing guidance; “a multi-stage ballistic missile” with a stated range goal of more than 2,000 kilometres; and a missile programme known as the “R2000” set, which included a hypersonic glide vehicle variant.
They used Claude Code “in place of human software engineers” to work on guidance, navigation and control software, according to the report.
Claude was relied upon for several aspects of the development process, including writing software, conducting research, tuning flight controls, running software builds and carrying out flight simulations.
Houthis ran multiple Claude sessions simultaneously, assigning different tasks to separate AI instances. One was used for coding, another for research and a third for reviewing the work produced by the first.
Anthropic said its safeguards blocked many of the group’s requests, but it found ways around some of those restrictions.
They attempted to conceal their objectives and the intended use of the software, while splitting their work across multiple sessions so that no single conversation revealed the full scope of the weapons programme.
Almost there
The group eventually carried out a live test of a guided rocket in Yemen, according to Anthropic.
The test appeared to fail. Within hours, the actors returned to Claude to investigate what had gone wrong, the company said.
Anthropic said it had no evidence that the group successfully fielded an operational weapon.
But the company found evidence of sustained efforts to develop and simulate weapons beyond the rocket programme.
Claude was also used in work on ballistic missile simulations and other weapons, including modelling trajectories and optimising flight-control systems, Anthropic said.
The actors also developed an offline simulation toolkit that did not rely on Claude or engineering software such as MATLAB. The toolkit was eventually compiled into a standalone programme that could “run and persist without Claude”.
That development allowed the group to continue its engineering work independently of Anthropic’s AI system, the company said.
Anthropic subsequently banned accounts associated with the actors and shared information about their activities with public- and private-sector partners.
The company said the case demonstrated the difficulty of preventing AI-assisted weapons development when users deliberately conceal their objectives and divide complex projects across multiple interactions.
The findings underscore concerns over how advanced AI systems could accelerate weapons development, particularly when users attempt to conceal their intentions and transfer work to tools that operate independently of AI platforms.
The news comes at a time when Houthis have ramped up attacks on Saudi Arabia’s oil installations, and Riyadh is responding with air strikes. The Iran-backed group has also taken control of Yemen’s strategic coastal city of Mocha as fighting continues with Yemen’s internationally recognised government.
