Anthropic says operators in northern Yemen used its Claude Code tool to perform work that would ordinarily require a specialized missile engineering team, according to the company’s September threat report. The cell developed software for guided rockets, a long-range ballistic missile exceeding 2,000 km range, and a hypersonic glide vehicle concept called “R2000.”
The operators ran multiple Claude instances simultaneously, dividing tasks among coding, research, and technical review functions. This parallel workflow allowed a small group to reproduce several functions normally distributed across specialist engineering teams.
The technical work extended into navigation, flight control, and simulation. According to Anthropic, the operators integrated open-source autopilot software with a phone-class flight computer and used Claude to write navigation and control code. They also developed six-degree-of-freedom trajectory simulations—a method for modeling an object’s movement and rotation through space—and used Claude for reinforcement learning work to tune flight-control algorithms. The operators compiled the project into a standalone executable that could run offline, enabling continued development without direct Claude access.
The activity moved beyond software development when the group test-fired a guided rocket in Yemen. According to Anthropic’s material, the test appears to have failed. Within hours, the operators returned to Claude and analyzed launch telemetry to investigate the failure. Anthropic said it found no evidence the group succeeded in fielding an operational weapon, but by the time accounts were disrupted, the operators had assembled an offline engineering toolkit.
Anthropric’s safeguards blocked numerous requests during the project. The operators adapted by obscuring the intended end use of individual tasks, dividing work across separate conversations, and employing other methods to circumvent restrictions. Anthropic subsequently banned accounts linked to the activity.
The company assessed the cell as highly likely to be linked to the Houthis. This operation marks what Anthropic describes as one of the clearest examples in its report of generative AI being applied directly to conventional weapons development. The wider threat report covers operations Anthropic said it disrupted between December 2025 and August 2026, including six conventional-weapons cases—three linked to China, two to Russia, and one to Yemen.
Key facts
- Operators ran multiple Claude Code instances in parallel to develop guidance software for guided rockets, ballistic missiles, and hypersonic glide vehicles
- The group developed six-degree-of-freedom trajectory simulations and compiled an offline-executable engineering toolkit
- A test-fired guided rocket appears to have failed, after which operators used Claude to analyze telemetry data
- Anthropic’s safeguards were circumvented through task obscuring and dividing work across separate conversations
- Anthropic found no evidence the group succeeded in fielding an operational weapon