Chinese Military Researchers Use US AI Models to Develop Defense Systems

A Reuters review of over 80 Chinese research papers and patents revealed that Chinese military researchers have employed outputs from advanced artificial intelligence models developed by the American companies OpenAI and Anthropic to train local AI systems intended to strengthen China's defense capabilities.

This previously undisclosed information highlights how Chinese military and security institutions leverage advanced US AI models to accelerate the development of their own specialized systems, despite Washington's efforts to restrict Beijing's access to advanced chips and other strategic technologies.

The documents demonstrate extensive use of a technique called "model distillation," where outputs from a sophisticated AI model are used to train smaller, more specialized models that can operate locally without requiring the massive computing power needed to build advanced systems from scratch.

The Reuters review, which included research collected by the Jamestown Foundation in Washington and shared exclusively with the agency, found that researchers linked to the People's Liberation Army and other military institutions widely apply model distillation technology.

The studies indicate that Chinese defense organizations view leading US AI models as a source of technical insight and a means to close the gap with American competitors.

The dispute centers on unauthorized data extraction rather than model distillation itself, which is a commonly used practice in the field.

This issue has emerged as a key point of contention ahead of US-China talks on AI governance and safety. US officials accuse certain Chinese entities of using model distillation to extract capabilities from American AI models, potentially undermining export controls and violating intellectual property rights.

China denies these accusations, asserting that Washington seeks to maintain "hegemony" over the AI sector and claims that American companies have engaged in similar practices.

Chinese developers reject claims that their AI advancements rely on foreign models.

A study published last year by researchers from Unit 96941 of the People's Liberation Army indicated the use of OpenAI's GPT-3.5 model to process sensitive military code. Unit 96941 is a military intelligence and electronic warfare unit.

The researchers stated that external models are unsuitable for handling classified information. To overcome this limitation, they used GPT-3.5 to summarize software code and then trained a local model on those summaries to operate entirely within Chinese military networks.

The White House, US Department of Defense, Chinese Foreign Ministry, People's Liberation Army, and OpenAI did not respond to requests for comment.