Chinese military-linked researchers are reportedly leveraging outputs from AI models developed by American companies like OpenAI and Anthropic to create smaller, domestically-produced AI systems. This process, known as knowledge distillation, allows them to utilize the reasoning and answers from these advanced models to train their own systems for various military applications, including surveillance and drone targeting. This strategy highlights a shift in the AI arms race, where the focus is now on maximizing military utility from existing technologies rather than solely developing cutting-edge models from scratch.

OpenAI: OpenAI develops advanced generative AI models including large language models used for text generation and reasoning tasks. Chinese military-linked researchers have drawn on outputs from these models to perform knowledge distillation, training smaller domestic systems for applications in surveillance, cyber operations, and battlefield decision-making. This approach enables efficient military AI deployment without building equivalent frontier models from scratch.
Anthropic: Anthropic creates AI models such as Claude that emphasize safety and reliable reasoning capabilities. Outputs and reasoning traces from these models have been used by Chinese researchers in distillation processes to build compact AI systems tailored for drone targeting, software analysis, and tactical military networks. The method leverages the models’ strengths while avoiding direct reliance on U.S. infrastructure.

Military AI Focus: The AI competition now emphasizes practical deployment of derived models on local hardware for defense uses rather than solely pursuing the largest models.
Distillation Method: Knowledge distillation transfers reasoning capabilities from large models to smaller ones by using generated answers and explanations as training data.