Chinese military researchers have leveraged outputs from U.S. artificial intelligence models developed by companies like OpenAI to train defense systems. This approach allows Chinese institutions to bypass the need for advanced chips and other restricted technologies, potentially accelerating the development of specialized AI systems for military use.

The use of U.S. AI models by Chinese researchers is part of a broader strategy to enhance China's technological capabilities. Sunny Cheung, a Jamestown fellow, analyzed over 60 papers and found that Chinese military scientists are systematically capturing the reasoning steps of Western models to adapt them for surveillance and other defense-related applications.

To overcome the limitations of training models from scratch, Chinese researchers have used GPT-3.5 to summarize software code and trained domestic models on those summaries. This technique allows models to run entirely within Chinese military networks, reducing reliance on foreign infrastructure and improving data security.

The use of U.S. AI models by Chinese institutions raises concerns about the cost, lock-in, and governance of AI technologies. It also highlights the potential for vendor lock-in and the challenges of maintaining control over AI systems that rely on foreign models. Market reactions suggest that such practices could influence global AI governance and export control policies.

The practice underscores the need for more robust international regulations to prevent the misuse of AI technologies. As the use of U.S. models by Chinese researchers continues, it may prompt further scrutiny of AI exports and the development of more secure, domestically controlled AI systems.