A weapons development cell operating in northern Yemen used Anthropic’s Claude AI to help build software for missile guidance, navigation and control, according to a new threat report by the US AI company.
Anthropic said the group was working on three weapons programs, including a multi-stage ballistic missile, a multi-variant missile and a guided rocket fitted with a phone-class flight computer.
The group used Claude Code to develop software for controlling and stabilizing the weapons. Its work included integrating an open-source autopilot, writing position-estimation and control software, adjusting control settings, building firmware and running flight simulations.
Anthropic said its safety systems blocked many requests but did not stop all of them. The group reportedly tried to bypass the safeguards by concealing the intended use of the software and dividing its work across multiple AI sessions.
The company said it later shut down the accounts involved and shared information about the activity with relevant partners.
The report also identified attempts to use Claude for biological weapons-related research, highlighting growing concerns over the use of advanced AI in weapons development.
