U.S. Treasury Secretary Scott Bessent said on July 21 that the administration would look at whether Chinese AI models were built by distilling American ones1, and warned that Chinese AI companies could face sanctions if they are found to have stolen intellectual property2.
"We've seen a lot of talk about open source models coming and threatening the large language models in the US," Bessent said on Fox Business. "This administration supports open source models, but what we do not support is IP theft. If we see, especially, that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft". Bloomberg was first to report the remarks.
Bessent also said the U.S. is finding watermarks of American large language models on many Chinese models and would be "looking at that in the coming days or weeks".
The threat comes as Chinese open-weight models gain on the leading American offerings from OpenAI and Anthropic. Chinese startup Moonshot AI put out its Kimi K3 model earlier in July. Washington has already restricted China's access to advanced chips and tightened export controls; sanctions would extend that pressure to the AI models themselves.
Axios reported on July 20 that the administration is weighing a wholesale ban on Chinese open source models, a claim others have disputed. The White House said in April that it would coordinate closely with AI companies to counter foreign efforts to copy their technology.
Bessent said the technical term for what he called theft is distillation, a training method in which a smaller model is built from the outputs of a stronger one. In June, Anthropic told the Senate Banking Committee in a letter that Alibaba had carried out the "largest known distillation attack" against it to date.
Whether distilling a rival's model amounts to theft is disputed. Microsoft CEO Satya Nadella said earlier in July that, while model providers' fair-use rights to train on public data are needed, he finds it "ironic that the status quo is to then turn around and impose restrictive terms on distillation". Hugging Face CEO Clem Delangue has called distillation "a very small factor in the ability to create good models" and a practice that "everyone is doing, including companies in the U.S."
Reuters has reported that the U.S. and China plan AI talks in September, with Bessent representing the U.S..
Anthropic and OpenAI have themselves been accused of theft: in 2023, The New York Times took OpenAI and Microsoft to court, claiming that training models on its work infringed its copyright.
ANALYSIS The threat targets Chinese labs while the American labs it would protect have faced accusations of their own, including a copyright suit over training data, leaving the line between permissible training and theft contested on both sides.