Newsroom
A cell believed to be linked to Yemen’s Houthi rebels used Claude AI to help develop the software needed to guide and control missiles, according to a new Anthropic threat report cited by the Financial Times.
In other words, the militants were not simply asking a chatbot general questions about weapons. They were reportedly using Claude Code as a substitute for human software engineers, tasking it with writing code, connecting an open-source autopilot to a flight computer, and simulating how a missile would behave in the air.
Anthropic did not directly name the Houthis. It described those involved as a group of “threat actors” operating in northern Yemen, an area largely controlled by the Iran-backed movement. The cell was reportedly working on three weapons programs involving a multistage ballistic missile, several missile variants, and a guided rocket.
According to the report, the group used an ordinary phone-grade computer as part of the rocket’s flight system. Claude allegedly helped write software to control the weapon’s movement, estimate its position, adjust its settings, and run virtual flight tests before launch.
The AI company said its safety systems stopped many of the requests, but not all of them.
Those behind the operation reportedly disguised what they were building and divided the work across several conversations, ensuring that no single session revealed the full project. It is the digital equivalent of handing different workers separate pieces of a blueprint so none sees the entire weapon.
There is no evidence that the group successfully produced an operational missile using Claude. Anthropic said a guided rocket was test-fired but appeared to fail. Within hours, those involved reportedly returned to the AI model to find out what had gone wrong.
Anthropic has since closed the accounts and shared information about the activity with government and private-sector partners. However, the company said the group had already created an offline simulation system that could continue operating without Claude.
The case offers a worrying glimpse of how AI is changing modern warfare. Skills that once required a team of experienced engineers, specialist software, and years of training can increasingly be supported by commercially available tools.
The danger is not that Claude independently designed and launched a missile. It is that AI may help a determined group move faster, solve technical problems, and fill gaps in expertise it does not already possess.
The same technology many people use to draft emails, explain homework, or fix computer code is now appearing inside alleged weapons programs.
*Source: Financial Times
