A cell of militants in northern Yemen used Anthropic’s Claude AI to help develop software for guided rockets and ballistic missiles, exposing a growing risk of advanced artificial intelligence being used to accelerate weapons development.
According to TRT World, Anthropic identified a “cell of threat actors” based in northern Yemen that used Claude Code as a substitute for human software engineers while working on three weapons programmes. The company did not identify the group as the Houthis, but northern Yemen is largely controlled by the Iran-aligned movement.
The cell used Claude to develop guidance, navigation and control software designed to steer and stabilise flying vehicles. The AI was also used for coding, technical research, flight-control tuning, software builds and flight simulations.
One of the projects involved a guided rocket equipped with a “phone-class flight computer” and final-stage homing guidance. The other programmes included a multi-stage ballistic missile with a stated range of more than 2,000 kilometres and a missile system known as the “R2000”, which included a hypersonic glide vehicle variant.
Financial Times reported that the group attempted to bypass Anthropic’s safeguards by concealing the purpose of its work and dividing the project across multiple Claude sessions. Separate AI sessions were reportedly used for coding, research and reviewing the work, making it harder for any single conversation to reveal the full scope of the weapons programme.
Anthropic said its safeguards blocked many of the group’s requests, but the actors found ways around some restrictions.
The group eventually conducted a live test of a guided rocket in Yemen, but the test apparently failed. Within hours, the actors returned to Claude to investigate the failure, according to Anthropic.
The company said it had no evidence that the group successfully deployed an operational weapon, although it found evidence of sustained efforts to develop and simulate guided weapons and ballistic missiles.
The actors also developed an offline simulation toolkit that could operate without Claude or conventional engineering software such as MATLAB. Anthropic said this meant the group could continue its weapons-related engineering work independently of the AI system.
Anthropic subsequently banned the accounts associated with the actors and shared information about their activities with public- and private-sector partners.
The case highlights a growing challenge for AI companies: preventing advanced models from being misused for military purposes when users deliberately conceal their intentions and divide complex projects across multiple interactions.
The report also raises wider concerns about the potential role of AI in accelerating missile and weapons development, particularly among armed groups with access to technical expertise and existing weapons programmes.