What is the story about?
A weapons development cell based in northern Yemen used Anthropic’s Claude Code to develop guidance software for guided weapons, according to a September
2026 report published by Anthropic. The company identified the activity during an internal investigation into suspected weapons development and said the actors used Claude in place of human software engineers. Anthropic said the cell was working on three weapons development programmes, including a guided rocket and ballistic missile projects. The investigation also identified a multi-variant missile programme referred to by the actors as the “R2000” set.
AI Used Across Weapons Development Work
According to Anthropic, the Houthis used Claude to develop software associated with the guidance, navigation and control of flying vehicles. The company said they used multiple Claude instances simultaneously, assigning different instances roles including coding, research and reviewing work produced by another instance.
The AI was also used in software development and simulation work, Anthropic said. The company said its safeguards blocked many of the actors’ requests, but not all of them.
Anthropic said the weapons developers used several methods to circumvent its safeguards. These included concealing their objectives and the intended use of the software, while splitting their activities across multiple sessions so that no individual interaction revealed the full purpose of the work.
The company said the activity represented a sustained effort to develop guided weapons. However, Anthropic stressed that it had no evidence the actors succeeded in fielding an operational device.
Guided Rocket Test Appears To Have Failed
The investigation did identify evidence that the group test-fired a guided rocket. Anthropic said the test appeared to have failed, after which the actors returned to Claude to investigate why the flight test had gone wrong.
The findings therefore do not establish that an AI-assisted operational missile was successfully developed or deployed. Instead, they show that a weapons development group was able to use a commercial AI coding system as part of its engineering effort.
Meanwhile, the AI company said it banned accounts associated with the actors after identifying the activity. The company also shared threat information with public- and private-sector partners to help mitigate the risks.
The investigation further found that the actors had already developed an offline simulation toolkit that did not depend on Claude or other engineering computing environments. This indicates that their broader weapons-development capability was not entirely dependent on access to Anthropic's AI systems.
