According to a report by TechCrunch, the United States is threatening to impose sanctions on Chinese artificial intelligence models due to suspicions of intellectual property theft. On Tuesday, US Treasury Secretary Scott Bessent officially stated that the United States will thoroughly examine open-source models originating from China to detect signs of intellectual property (IP) theft, and even threatened direct sanctions against Chinese AI companies if such IP theft is proven. These remarks come against the backdrop of the rapid advancement of Chinese models, led by Moonshot AI’s Kimi K3 model, which present enhanced capabilities and growing popularity. This progress threatens to harm the business models of leading AI companies in the United States, such as OpenAI and Anthropic, as well as their ability to raise additional capital to continue developing frontier models.
Treasury Secretary Scott Bessent's Statement on Fox Business
During an interview with Fox Business on Tuesday, whose remarks were first reported by the Bloomberg news agency, US Treasury Secretary Scott Bessent addressed the growing concerns in Washington regarding Chinese developments in the open-source sector. Bessent stated in the interview: "We’ve seen a lot of talk about open source models coming and threatening the large language models in the US." He added and emphasized the current administration's position, saying: "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." This statement marks a new phase in the US administration's efforts to protect the intellectual property of leading domestic companies.
Concern Over Damage to the Business Models of OpenAI and Anthropic
The strengthening of Chinese models, spearheaded by Moonshot AI's Kimi K3, poses a direct threat to the business models of leading American companies like OpenAI and Anthropic. This disruption could make it difficult for these companies to raise the capital required to continue developing advanced frontier models that require massive resources. Concurrently, AI companies in the United States have been warning for many months about campaigns led by foreign actors aimed at copying these companies' AI technology and subsequently redeploying it to the public as open-source models. Back in April, the White House announced it would work closely with AI firms to combat this technology theft.
Reports on Trump Administration's Intention to Impose a Wholesale Ban on Chinese Models
In addition to Treasury Secretary Bessent's statements, on Monday, the Axios website reported that the Trump administration is considering imposing a wholesale ban on Chinese open-source models, although other industry figures disputed this claim. The potential imposition of sanctions by the United States against Chinese models would join a growing list of strategies through which the administration in Washington is trying to maintain American leadership in the global AI race. This step comes after the United States already restricted China's access to advanced chips and tightened export controls on them. Now, Washington is signaling that it may act directly against the AI models themselves—a move that could mark a significant escalation in the technological competition between frontier model labs and Chinese open-source alternatives.
The Model Distillation Technique and the Controversy Over Its Definition as Theft
One of the core tools at the heart of the technological and legal debate is "Model distillation." Model distillation is a technique that allows translating some of the capabilities of a larger model into a smaller system, which is easier and more convenient to run. However, not all industry participants agree that distilling a model belonging to another company indeed constitutes intellectual property theft.
Earlier this month, Microsoft CEO Satya Nadella criticized the large AI labs that make this assumption. Nadella said: "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."
Is Distillation the Reason for Chinese Success? Industry Expert Opinions
Some players in the AI industry argue that model distillation is not the only factor, and perhaps not even the primary one, in China narrowing the gap with American companies. In a recent episode of TechCrunch’s Equity podcast, Hugging Face CEO Clem Delangue addressed this issue. Delangue said: "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." He added: "If it were easy just to do distillation to get good at building AI models, there would be many other countries, including in the U.S., with much better open source AI." Delangue further noted that "The reality is they have really, really good research teams in China… taking a much more open and collaborative approach to AI than in the U.S."
Legal Risks and Model Training Practices in the United States
In parallel to international disputes, the training practices of leading AI labs continue to represent a significant source of legal risk for companies in the United States itself. This week, Anthropic received approval to start sending checks to authors as part of a $1.5 billion settlement. This settlement was reached after a judge ruled that the company had illegally downloaded and stored millions of copyrighted books to train its AI models, highlighting the legal complexities surrounding copyright and model training in this field.