Chinese military researchers tap US AI models to train defense systems

Chinese military researchers have used outputs from OpenAI and Anthropic models to train domestic AI systems for defense, according to a Reuters review of more than 80 Chinese academic papers and patents. The documents reviewed show widespread use of model distillation, a method in which outputs from a large model are used as training material for smaller, specialised models.

The Reuters review covered more than 80 Chinese academic papers and patents. It found that researchers working on defense-related AI systems drew on outputs from models developed by OpenAI and Anthropic. According to the article, the use of distillation to train smaller specialised models was widespread across the documents examined. Distillation is a common AI technique that allows a smaller model to be created without the computing expense of full-scale training, using the larger model's outputs as a source of the larger model's capability.

The excerpt does not identify specific defense systems or applications, and it does not include comment from OpenAI, Anthropic, or Chinese officials. The report is based on a review of published papers and patents rather than direct observation of military programs. The article also does not say whether the papers and patents in question were prepared in collaboration with, or solely for, military agencies. The findings highlight a pathway by which outputs from US AI models can be incorporated into foreign defense research.