Anthropic Reports AI Misuse in Yemen Missile Programs

AI developer Anthropic reported that a weapons cell in northern Yemen utilized its Claude AI model to assist in the development of guidance, navigation, and control software. The misuse targeted three specific weapons programs, including a multistage ballistic missile and a hypersonic-glide vehicle. In a separate incident, a suspected actor linked to Iran used the same model to generate targeting intelligence on U.S. naval forces by analyzing ship and satellite data.

Anthropic stated it found no evidence that operational weapons were deployed based on this AI assistance. However, the discoveries prompted CEOs Dario Amodei and Sam Altman to call for a slower development pace for frontier models. They advocated for the implementation of common safety standards and the use of independent oversight to prevent future misuse.

These security breaches occur amid increasing competitive pressure from Chinese developers. Open-weight models from firms such as DeepSeek, Zhipu, and MiniMax are rapidly closing the capability gap with closed U.S. models, a trend that may undermine the effectiveness of unilateral safety restrictions imposed by American AI labs.