China’s military and security-linked institutions are increasingly using outputs from leading US AI models to develop specialised domestic systems, according to a review of more than 80 Chinese academic papers and patents. The findings indicate that researchers linked to the People’s Liberation Army (PLA) are using model distillation to adapt capabilities from advanced US systems for local defence applications.
Model distillation involves using outputs from a powerful AI model to train a smaller, specialised model that can operate locally with lower computing requirements. The research reviewed indicates that the method is being used by PLA-linked institutions and other military organisations.
The documents suggest Chinese defence researchers view leading US models as a source of technical knowledge and a way to narrow the gap with American rivals. The concern is focused on unauthorised extraction rather than distillation itself, which is widely used across the AI industry.
The issue is gaining importance ahead of US-China discussions on AI governance and safety. US officials have accused some Chinese entities of using distillation to extract capabilities from American models, potentially bypassing export controls and raising intellectual property concerns. China has rejected the allegations, accusing Washington of AI “hegemonism” and arguing that US companies have used similar practices.
Sunny Cheung, a Jamestown fellow who examined more than 60 papers, said Chinese military scientists are capturing reasoning processes from Western AI models for surveillance, cyber warfare and tactical decision-making.
“Teaching a model the right answer is one thing but teaching it the reasoning behind the answer is much harder,” Cheung said.
A PLA Unit 96941 paper described using OpenAI’s GPT-3.5 to summarise sensitive military source code before training a domestic model on those summaries. The system could then operate within Chinese military networks.
Other research included using Anthropic’s Claude 3 Haiku to generate training data for social media monitoring, distilling an image-processing model for drones and deploying target-recognition systems on tactical hardware during simulated maritime operations.
Anthropic said it does not provide commercial access to Claude in China or Beijing-controlled firms. It also warned that distilled models may lose safety safeguards.
China is promoting model lightweighting and edge computing amid US restrictions on advanced chips. Researchers are also studying “data-free distillation” as a security threat.
Experts caution that distilled systems inherit selected capabilities rather than the full intelligence of frontier models.
“It is best understood as transferring selected capabilities into a cheaper, locally controlled system, not achieving independence from frontier AI.”
Also read: Viksit Workforce for a Viksit Bharat
Do Follow: The Mainstream LinkedIn | The Mainstream Facebook | The Mainstream Youtube | The Mainstream Twitter
About us:
The Mainstream is a premier platform delivering the latest updates and informed perspectives across the technology business and cyber landscape. Built on research-driven, thought leadership and original intellectual property, The Mainstream also curates summits & conferences that convene decision makers to explore how technology reshapes industries and leadership. With a growing presence in India and globally across the Middle East, Africa, ASEAN, the USA, the UK and Australia, The Mainstream carries a vision to bring the latest happenings and insights to 8.2 billion people and to place technology at the centre of conversation for leaders navigating the future.