Chinese military research units are allegedly using US AI models to train their smaller, portable models for defense systems, according to findings released on Friday. Researchers reviewed over eighty Chinese academic papers and patents, finding that Chinese units used a technique called "model distillation," which involves using a more powerful AI system to train smaller, specialized models deployable without major computing power. Chinese defence institutions view these frontier models as a source of technical insight and a way to close the AI gap with the US. Sunny Cheung, a Jamestown fellow who analysed over sixty papers, noted that Chinese military-linked researchers are trying to transfer proprietary reasoning from Western models into smaller systems. In one paper by PLA Unit 96941, GPT-3.5 was used to process sensitive military source code, with researchers emphasizing that third-party models were unsuitable for classified information. This research comes as the US and China plan to meet in September for AI talks to discuss mitigating risks posed by frontier AI models. Treasury Secretary Scott Bessent commented, "We are finding watermarks of our US large language models on many of the Chinese models, and that’s unacceptable." The US may strengthen export controls on AI model diffusion; the former Biden administration implemented the AI diffusion framework in January 2025, but the Trump administration undid these rules in May 2025, stating the requirements would have "stifled American innovation and saddled companies with burdensome new regulatory requirements."