Academic papers describe model distillation for defence and security applications
BEIJING: Chinese military-linked researchers have used outputs from leading US artificial intelligence models developed by OpenAI and Anthropic to help train domestic AI systems, according to a Reuters review of more than 80 Chinese academic papers and patent filings.
Reuters reported that the documents provide new insight into how Chinese defence and security institutions are using a technique known as “model distillation” to build smaller, specialised AI models. The approach involves using the outputs of advanced AI systems to train more compact models that require significantly less computing power to operate.
According to Reuters, researchers affiliated with the People’s Liberation Army (PLA) and other military institutions have widely adopted the technique despite ongoing US export restrictions on advanced chips and AI technologies.
Papers highlight military applications
The Reuters review, which included research compiled by the Washington-based Jamestown Foundation, found that Chinese researchers have explored using distilled AI models for surveillance, cyber operations, drone navigation, target recognition and battlefield decision-making.
One paper reportedly 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 within Chinese military networks. Other studies cited by Reuters involved image-processing systems for unmanned aerial vehicles and AI models designed for maritime operations.
Anthropic said it does not provide commercial access to its Claude models in China or to entities controlled by Beijing and uses monitoring systems to detect policy violations.
Governance debate continues
Reuters noted that the findings come as AI governance and security remain major issues in US-China relations. US officials have accused some Chinese organisations of using model distillation to extract capabilities from American AI systems, potentially undermining export controls and intellectual property protections.
China has rejected those accusations, arguing that Washington is pursuing what it describes as AI “hegemonism.” Chinese AI companies have also denied claims that their technological progress relies primarily on foreign models.
Neither the White House, the Pentagon, China’s Foreign Ministry, the PLA nor OpenAI immediately responded to Reuters’ requests for comment.
Experts quoted by Reuters cautioned that distilled models remain less capable than the original frontier AI systems and transfer only selected capabilities rather than replicating their full performance.
