US Treasury Secretary Scott Bessent said that the Trump administration will investigate whether Chinese AI models were built by distilling American ones, and warned that companies found to have done so could face sanctions.
Speaking on Fox Business, Bessent said the administration supports open-source models but not intellectual property theft, adding that if overseas models are “stealing from our great companies, we have the ability to sanction them.” He claimed US officials are finding “watermarks” of American large language models inside many Chinese models, calling it unacceptable and saying the government would examine the issue in the coming days or weeks. He offered no public evidence for the watermark claim.
Distillation is the practice of training a model using the outputs of a larger, more capable system, which can transfer much of the stronger model’s capability at a fraction of the training cost. The technique is widely used and legitimate in many contexts; the dispute is over scale and consent.
Anthropic wrote to the Senate Banking Committee last month alleging that Alibaba had conducted what it called the largest known distillation attack against it, and White House technology director Michael Kratsios separately accused Moonshot AI of distilling Anthropic’s Fable model to build Kimi K3. Moonshot has not publicly responded, and Beijing has dismissed such accusations as slander. The threat carries diplomatic weight because Bessent is set to lead the US delegation at the first bilateral AI talks with China in September.
What makes enforcement complicated is how deeply Chinese models are already embedded in American development. On OpenRouter, the routing platform many developers use to access models, Chinese systems accounted for roughly 58% to 63% of tokens used by US firms in recent weeks, depending on the source and period, up from under 10% a year ago. DeepSeek alone is the platform’s largest single vendor. A year ago US-built models handled around 70% of that traffic; Bloomberg reports the figure is now closer to 30%. One startup said switching from Anthropic’s models to DeepSeek cut its inference costs by about 90%.
The Enforcement Problem
Restricting Chinese models would be far harder than earlier crackdowns on Chinese electric vehicles or Huawei equipment. Open weights are already downloaded and running on American servers, so there is no shipment to block at a border and no vendor relationship to sever.
Officials have floated tools including financial sanctions on designated companies or orders requiring US firms and cloud providers to audit and remove models containing allegedly stolen weights, but each would fall directly on American businesses that have built products on this software.
That is precisely the objection nearly 200 startups raised in letters urging the administration not to cut off access, and even inside the White House, adviser David Sacks has warned against regulatory capture by large US labs. The lever Washington is reaching for lands hardest on its own developers.
A Charge That Cuts Both Ways
The intellectual property framing is complicated by the American labs’ own records. Anthropic agreed to pay $1.5 billion to settle a class action brought by authors who alleged it downloaded books from pirated databases, a settlement a federal judge approved this week. The New York Times sued OpenAI and Microsoft in 2023 over the use of its journalism in training data, and similar suits are pending against Google and Meta.
That history does not resolve whether Chinese labs distilled American models without permission, a factual question that remains unproven in public. But it does mean the argument arrives on contested ground, since the same companies asking Washington to police how their outputs are used are themselves defending how they acquired their inputs. Whether the administration acts before the September talks, or lets the threat serve as leverage, is the immediate question.
Disclaimer: AIstify is an independent media brand owned and operated by NuvexMedia LLC, publishing news, research, and insights on artificial intelligence, emerging technologies, automation, and related industries. NuvexMedia LLC invests in and collaborates with companies across the AI, technology, software, and digital innovation sectors. These relationships do not influence AIstify’s editorial coverage, and the publication maintains full editorial independence to provide accurate, timely, and objective information. © 2026 NuvexMedia LLC. All rights reserved. This content is for informational purposes only and should not be considered legal, tax, investment, financial, or other professional advice.
