Anthropic identifies AI use by group in Yemen for missile projects
Company claims that group linked to the Houthis used the Claude chatbot to try to develop weapons guidance systems.
Quick Look
- Anthropic revealed that a group based in northern Yemen, controlled by the Houthis, used the Claude chatbot to try to develop ballistic missiles and guided rockets.
- There is no evidence that the group created operational weapons.
AI-generated summary
Why It Matters
The Houthis are a political and military movement in Yemen, supported by Iran, involved in regional conflicts and attacks on ships.
Among them were a guided rocket, a ballistic missile with a range of more than 2,000 kilometers and a projectile equipped with a hypersonic glider. Despite the progress of the projects, Anthropic said it found no evidence that the group had managed to put an operational weapon into operation.
According to the company, the group turned to Claude to carry out tasks that would normally require a team of engineers specializing in guidance, navigation and weapons control systems.
One of the projects involved a rocket capable of automatically adjusting its trajectory during flight. To achieve this, the group developed a guidance system that used a flight calculator similar to that of a smartphone.
In practice, this type of system allows the equipment to calculate its position and make adjustments to try to follow a determined route to the target.
Anthropic stated that the group even tested the launch of this rocket. The attempt, however, failed.
Ballistic missiles are weapons that, after launch, travel a large part of their trajectory in a high-altitude curve before heading towards the target. The range of more than 2 thousand kilometers would allow it to reach regions very far from the launch site.
The third project mentioned by Anthropic involved a hypersonic glider. Unlike an object that simply follows a predictable trajectory, this type of vehicle can perform maneuvers during flight and reach speeds greater than five times the speed of sound.
Despite the projects, Anthropic said there is no evidence that the group was able to turn the work into an operational weapon.
The company also highlighted that one known test of a guided rocket ended in failure.
The report is part of a broader investigation by Anthropic into illegal or malicious uses of its artificial intelligence system. The company stated that it had interrupted operations involving users from different countries.
Anthropic has not publicly identified the group responsible for the projects. The company said only that he was based in northern Yemen, a region controlled by the Houthis.
The Houthis are an Iranian-backed political and military movement that controls much of northern Yemen and the country's capital, Sana'a. The group is involved in the Yemeni civil war and also became known internationally for attacks carried out against ships in the Red Sea and against Israel.
Open Questions
- What specific measures has Anthropic taken to block access?
- Was there collaboration with intelligence agencies?






