Chinese Military Trains AI Systems Using OpenAI and Anthropic Models

A Reuters investigation of more than 80 academic papers and patent documents found that institutions linked to the Chinese military have been using outputs from Western artificial‑intelligence models as synthetic training data for domestically developed AI systems. The study identified the use of large‑language‑model outputs from OpenAI and Anthropic in a process known as model distillation, where responses from advanced models are repurposed to train smaller, locally‑controlled models. Researchers at the Washington‑based Jamestown Foundation, including analyst Sunny Cheung, said the documents show attempts to transfer the reasoning capabilities of Western AI into Chinese defense‑related projects, enabling cost‑effective development of AI that can operate without reliance on foreign hardware.

The findings highlight a growing trend of leveraging commercially available AI technology for military research, raising concerns about the diffusion of advanced AI capabilities into defense applications and the potential for an AI arms race.