Researchers linked to China’s military have used outputs from leading United States artificial intelligence models to develop domestic defence-related AI systems, according to a Reuters review of more than 80 Chinese academic papers and patents. The findings suggest Chinese military AI projects are using a technique known as model distillation to build specialised systems while reducing reliance on large computing resources.
According to the review, researchers associated with the People’s Liberation Army (PLA) and other defence institutions have relied on outputs from AI models developed by OpenAI and Anthropic. Instead of copying the models themselves, the studies describe using their responses to train smaller AI models capable of operating within local military networks.
The documents indicate that model distillation allows developers to transfer selected capabilities from advanced AI systems into lightweight domestic models. This approach enables specialised applications while avoiding the significant computing power required to build frontier AI models from the ground up. Reuters said the reviewed papers portray leading US AI models as valuable technical references for Chinese military AI development.
One paper cited by Reuters described researchers from PLA Unit 96941 using OpenAI’s GPT-3.5 to summarise military software code before training a domestic model capable of operating entirely inside Chinese military infrastructure. The researchers reportedly argued that third-party AI models were unsuitable for directly handling classified information.
Other studies showed defence-related applications ranging from surveillance and social media monitoring to unmanned aerial vehicles and maritime operations. Researchers at the PLA’s National University of Defence Technology reportedly used model distillation to optimise image-processing systems for drones, while China’s Academy of Military Sciences tested similar methods for target recognition during simulated naval operations.
The findings emerge amid increasing competition between Washington and Beijing over artificial intelligence. US officials have expressed concerns that some Chinese entities may be using model distillation to extract capabilities from American AI systems, potentially weakening export controls and raising intellectual property issues. Chinese officials and some domestic AI companies have rejected those allegations, arguing that innovation is based on independent research.
Experts quoted by Reuters said model distillation can transfer selected capabilities but cannot fully reproduce the broad intelligence of advanced frontier AI systems. Neither the White House, the Pentagon, China’s Foreign Ministry, the PLA, nor OpenAI had publicly commented on the Reuters findings at the time of publication. The report highlights the growing strategic importance of Chinese military AI as global competition over artificial intelligence continues to intensify.
