US Treasury Secretary Scott Bessent on Tuesday threatened to sanction Chinese AI models over alleged intellectual property theft, signaling a major escalation in Washington’s efforts to maintain its lead in artificial intelligence.
Speaking on Fox Business, Bessent said the administration would examine open source models from China for signs of IP theft. “We’ve seen a lot of talk about open source models coming and threatening the large language models in the U.S.,” he said. “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.”
The comments, first reported by Bloomberg, come as Chinese AI models gain capabilities and popularity. Moonshot AI’s Kimi K3 was cited as a recent example of Chinese advancement. The rise of these models threatens the business models of top US firms like OpenAI and Anthropic, and their ability to raise capital for frontier models. Bessent’s remarks mark the first time a senior US official has explicitly threatened sanctions against AI models themselves, rather than hardware or chip exports.
A New Front in the Tech Race
The threat of sanctions adds to a growing list of strategies the US is deploying to stay ahead. Washington has already restricted China’s access to advanced chips and tightened export controls. Now it is signaling it may target AI models themselves, marking a significant escalation in the technological competition between frontier labs and Chinese open source alternatives.
The Axios news outlet reported Monday that the Trump administration is considering a wholesale ban on Chinese open source models. Others have disputed that claim. AI companies have warned for months about foreign actors copying AI tech and redeploying it as open source. The White House said in April it would work closely with AI firms to combat theft. Bessent’s Fox Business interview was the first time a Treasury secretary has directly linked sanctions authority to AI model theft, a move that could reshape how open source AI is distributed globally.
The Distillation Debate
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
At the heart of the dispute is a technique called model distillation, which translates a larger model’s capabilities into a smaller system. Not everyone agrees that distilling another company’s model constitutes theft. The practice is common in the industry, with many US companies also using it to improve their own models.
Microsoft CEO Satya Nadella criticized large labs for assuming distillation is theft. “While the great innovation that comes from model providers having fair use rights to train models on public data is needed, I find it ironic that the status quo is to then turn around and impose restrictive terms on distillation,” Nadella said. His comments highlight a tension within the US tech industry, where companies that rely on public data for training are now seeking to restrict how others use their outputs.
Some in the industry argue distillation is not the only reason China is catching up. Hugging Face CEO Clem Delangue said distillation is a minor factor. “We know distillation to be a very small factor in the ability to create good models, and it’s a practice that everyone is doing, including companies in the U.S.,” Delangue said. He added that if distillation were easy, many other countries would have better open source AI. “China has really good research teams taking a more open and collaborative approach than the US,” Delangue said. His comments suggest that China’s progress stems from genuine research capability, not just copying.
Legal Risks for US AI Labs
AI labs’ training practices continue to be a source of legal risk. This week, Anthropic got the green light to start cutting authors checks as part of a $1.5 billion settlement. A judge ruled that Anthropic illegally downloaded and stored millions of copyrighted books to train its AI. The settlement is one of the largest in the AI industry to date and sets a precedent for how training data disputes may be resolved.
The settlement highlights the legal gray areas surrounding AI training data, even as the US government threatens foreign competitors over similar practices. Critics note that US companies have themselves been accused of using copyrighted material without permission, raising questions about the consistency of Washington’s stance on IP theft in AI. The contrast between the Treasury’s threats against China and the legal settlements at home underscores the complexity of enforcing intellectual property rules in a rapidly evolving field.