The Trump administration is preparing to examine Chinese AI models for evidence of intellectual property theft, with Treasury Secretary Scott Bessent warning that companies found to have copied American AI technology could face US sanctions.
Bessent said the administration supports open-source AI development but drew a distinction between open innovation and the unauthorized use of proprietary American technology. He said the government is reviewing whether Chinese AI developers have relied on US models to accelerate development of their own systems.
Chinese AI developers have narrowed the performance gap with leading US model providers, prompting debate over whether American companies can maintain their technological lead. The latest catalyst was the release of Kimi K3 from Chinese startup Moonshot AI, an open-weight model that has drawn attention for posting strong results on several industry benchmarks. Additionally, Alibaba released Qwen 3.8, claiming it is “second only to Fable” in performance.
The emergence of competitive Chinese models has unsettled investors, suggesting that Silicon Valley’s multibillion-dollar investment in AI infrastructure may face growing pressure from lower-cost competitors.
AI Distillation
The administration’s concern is a technique known as distillation, in which a smaller AI model is trained using the outputs generated by a larger, more advanced model. While distillation has legitimate uses within AI development, US AI companies argue that it can also be used improperly to replicate proprietary capabilities without access to the original training process.
OpenAI and Anthropic have both alleged that Chinese competitors have used outputs from their models without authorization to develop rival systems more quickly and at substantially lower cost. Earlier this year, the Trump administration pledged additional measures to hold foreign organizations accountable for such practices.
Bessent said the administration has identified what he described as watermarks from US LLMs appearing in several Chinese AI systems. He did not identify specific companies or provide technical evidence supporting the claim but said the findings would be reviewed over the coming weeks.
The Treasury secretary also suggested another possible policy response beyond sanctions. Companies deploying Chinese AI models could eventually be required to disclose that information to customers, an approach he compared to identifying counterfeit goods in commercial markets.
The administration’s scrutiny extends beyond the Treasury Department. US Trade Representative Jamieson Greer said officials are also evaluating whether China is using unfair competitive practices to strengthen its position in AI. He said the goal is to ensure American companies compete on equal terms in the global AI market.
US-China Talks Scheduled
The issue is expected to feature prominently in upcoming discussions between Washington and Beijing. US and Chinese officials plan to hold AI-focused talks in September, with Bessent expected to participate as the two countries continue negotiations on trade and tech policy.
Even as Bessent alleges that China has copied American AI tech, US AI developers face legal scrutiny over their sourcing of training data. Anthropic recently agreed to pay $1.5 billion to settle litigation brought by authors over copyrighted books.