Anthropic has revealed that Yemen’s Ansar Allah recently used its Claude AI models to develop software involved in guiding, navigating and controlling missiles and a guided rocket, according to the company’s latest threat intelligence report.

What Anthropic Found

The report identifies Ansar Allah as operating a cell running three distinct weapons programs, including a guided rocket, a multi-stage ballistic missile program with a stated range goal of more than 2,000 kilometers, and a separate multi-variant missile initiative. Anthropic said the group used Claude Code, its coding-focused AI tool, to develop software for controlling and stabilizing flying vehicles, reportedly integrating an open-source autopilot system with a phone-class flight computer and running flight simulations, and even test-firing a guided rocket as part of the process. Anthropic said its safety systems blocked many requests, but the group disguised the software’s purpose and split development work across multiple sessions to evade detection.

Anthropic’s Response and Why It Matters

Once the activity was identified, Anthropic said it banned the associated accounts and shared relevant information with public and private sector partners. The case illustrates a growing concern among AI safety researchers and policymakers that increasingly capable AI coding tools, built for legitimate software development, can be misused by armed groups to accelerate technical work that would otherwise require specialized engineering expertise, a concern likely to keep AI misuse monitoring a key focus area going forward.