Despite US Restrictions: China's Military Continues to Use American AI

A Reuters investigation revealed that China's military continues to use American AI models, including those from OpenAI and Anthropic, despite Washington's government restrictions. Chinese researchers use 'model distillation' to train local systems, but experts warn that this method transfers only part of the capabilities and does not provide full independence.

A new investigation by Reuters news agency revealed that China's military continues to use American AI models, including those from OpenAI (ChatGPT) and Anthropic (Claude), despite attempts by the Washington government to restrict their use. Reuters surveyed more than 80 Chinese academic papers and patents and found that Chinese military and security institutions use outputs from these models to train local AI systems and enhance military capabilities. Chinese researchers use 'model distillation', a technique that allows training smaller, more specialized models using outputs from powerful models, without the need for massive computing power. However, experts emphasize that this technology transfers only part of the original models' capabilities and does not provide full independence, limiting the use in tools like drones, satellites, and other devices. To address this limitation, researchers at the Military Engineering University published a study on 'data-free distillation', which allows reverse-engineering model capabilities without direct access to parameters. But this solution is also risky, as it may expose logical information that the models generated, which China does not intend to allow.