Anthropic Report: Yemen Cell Used Claude Code to Develop Software for Long-Range Missiles
Anthropic’s new threat report says a cell in northern Yemen, assessed as highly likely linked to the Houthis, used its Claude Code product in place of a full missile engineering team to build guidance software for a tactical rocket, a ballistic missile with more than 2,000 km range, and a hypersonic glide vehicle variant.
A new threat assessment from Anthropic describes a weapons project that leaned heavily on its own AI coding tool.
According to the report, a cell operating in northern Yemen and assessed as highly likely to be Houthi‑linked used Claude Code as a substitute for an entire missile engineering team. The group reportedly developed guidance software for three types of systems: a tactical guided rocket, a ballistic missile with a range beyond 2,000 kilometers, and a hypersonic glide vehicle variant.
The report says the operatives didn’t rely on a single chat session. Instead, they ran multiple Claude instances in parallel, with one focused on writing code, another on research, and a third on reviewing and refining outputs. That setup allowed them to iterate more quickly on control algorithms and software architecture than they could have with limited human expertise alone.
Anthropic’s account does not name specific missile designs, but the ranges described would, in principle, extend the reach of launchers in northern Yemen well beyond their immediate neighborhood.
The group in question operates in a context where Houthi forces have already used missiles and drones in real‑world attacks on Red Sea shipping and targets in Saudi Arabia and the United Arab Emirates. Any improvement in their ability to design and adapt guidance software for longer‑range systems directly affects the range of potential targets.
From a technical standpoint, the case highlights how generative AI tools can compress the time and expertise needed to produce complex code for weapons systems, especially for actors who already have some baseline technical know‑how but lack a full engineering team.
For governments and regulators, the episode turns a broad concern into a specific scenario: off‑the‑shelf AI services, accessed over the internet, being used to accelerate missile guidance software development by a non‑state group tied to an ongoing regional conflict.
The report lands as regional and U.S. officials are already wrestling with Houthi attacks on shipping and the broader question of how to contain them. A separate Axios report says Saudi Crown Prince Mohammed bin Salman recently urged former U.S. President Donald Trump to strike Houthi targets, and that the head of U.S. Central Command flew to Riyadh for emergency coordination.
Key signals to watch are whether technology firms tighten monitoring and access for high‑risk use patterns, how quickly governments move to fold AI services into export‑control and sanctions regimes, and whether other cases emerge of armed groups using similar AI‑driven workflows for weapons development.
Sources
- OSINT