- Chinese military researchers have used outputs from U.S. AI models like OpenAI's GPT-3.5 and Anthropic's Claude to train domestic defense AI systems via a technique called model distillation.
- A Reuters review of over 80 Chinese academic papers and patents reveals widespread use of distillation by PLA-linked institutions to develop specialized AI for surveillance, cyber warfare, and tactical decision-making.
- Model distillation involves using outputs from a powerful AI system to train smaller, specialized models, reducing the computing requirements compared to building frontier AI from scratch.
- The practice is disputed, with U.S. officials accusing Chinese entities of unauthorized extraction that undermines export controls and infringes IP, while China denies reliance and accuses the U.S. of 'hegemonism'.