A weapons development cell in northern Yemen reportedly used Anthropic’s Claude AI to work on missile guidance and control software, according to a new threat intelligence report from the AI company. Anthropic said the activity was part of a sustained effort involving several weapons programs, raising fresh concerns about the growing use of advanced AI in military applications.
The company did not publicly identify the group behind the activity. However, the location and other details described in the report have led outside observers to associate the operation with Yemen’s Iran-backed Houthi movement. Anthropic itself has not confirmed that identification.
According to Anthropic, the Yemen-based actors used Claude Code as a substitute for human software engineers while working on guidance, navigation, and control systems for flying weapons. The AI reportedly supported software development, simulation, and related engineering tasks.
Anthropic said the group was working on multiple weapons programs, including a guided rocket and missile projects. The company stressed that it had no evidence that the actors successfully fielded an operational weapon. Anthropic said the test appeared to fail, after which the operators returned to Claude to investigate the problem.
Anthropic said its safety systems blocked many of the actors' requests. The users nevertheless attempted to work around those safeguards by concealing the intended purpose of their work and dividing their activities across different sessions.
That approach made it harder for individual interactions to reveal the project's full context. The company said it subsequently banned accounts associated with the activity and shared relevant intelligence with public- and private-sector partners.
Also Read: ‘They’re Gambling With Our Lives’: Anthropic Researcher Sounds AI Alarm
The Yemen case was one of several examples outlined in Anthropic’s latest threat report. The company said it identified six operations involving the misuse of Claude for conventional weapons development, intelligence gathering or procurement across Yemen, China and Russia.
The report also documented attempts to use Claude in biological research, cyber operations, surveillance and influence campaigns. Anthropic said the incidents demonstrate how increasingly capable AI systems can be repurposed for harmful activities despite existing safeguards.
The findings add to the wider debate over how AI companies should monitor powerful models and prevent their use by organizations seeking to develop weapons or conduct other dangerous activities. Anthropic said it is strengthening safeguards and threat monitoring as its models' capabilities continue to advance.