White House aides have prepared to confront Beijing over the alleged theft of U.S. artificial intelligence models by their Chinese competitors, in what would be an escalation of the United States’ posture toward its technological rivals.
With Chinese President Xi Jinping set to visit this week, White House officials have readied evidence and talking points aimed at defending domestic AI giants from what they characterize as Chinese violations of international trade law, according to three sources familiar with the matter.
The preparations, not previously reported, come amid a broader debate among Trump advisers about the extent to which the U.S. should fight back against “open-source” AI models primarily offered by Chinese companies.
Some White House officials and many industry experts see the open-source models as undermining American economic and technological supremacy in the field, and want the U.S. to ask China to crack down on their proliferation.
Trending
But not every American firm wants to see limits imposed on the Chinese open-source models — including some leaders close to the Trump administration. Former Trump crypto czar David Sacks and Nvidia CEO Jensen Huang, among others influential with the president, have publicly touted the potential benefits of open-source models, which typically offer a free or far cheaper product and could accelerate technological adoption by American firms.
The American AI companies have themselves been accused of massive intellectual property theft, having been built on vast troves of data scraped from the Internet without creators’ consent.
The new debate centers on what is known in AI parlance as “distillation,” or the transfer of knowledge from a large AI model to a smaller one. While some versions of this practice are broadly accepted, the U.S. Cybersecurity and Infrastructure Security Agency publicly accused Chinese AI firms this month of “aggressive, malicious, and targeted” distillation at “an industrial scale.”
Huang signed on to a public letter in July that warned that open-weight models “are an important part of that foundation because they make advanced AI more accessible, adaptable, and widely available.”
Similarly, Sacks has maintained that “restrictions that kneecap open models would do nothing to stop Chinese labs” from creating the next iteration, but would “cripple the ability of startups” in the U.S. to create high-performance, low-cost models. He added, “Of course some of the closed labs would love this — it eliminates their competition.”
The administration already has a long list of priorities it plans to discuss with the Chinese on everything from tariffs to rare earth metals, and it’s unclear how much the meetings will focus on AI. Several sources familiar with the talks also downplayed the possibility for any breakthrough on the issue of “distillation.” Even before the rise of AI, past administrations have struggled to persuade China to limit alleged intellectual property theft. Doing so is considered even more unlikely for AI given its central importance.
A White House spokesperson declined to comment. The sources familiar with the administration’s preparations spoke to The Washington Sun on the condition of anonymity to reflect private deliberations.
“Getting China to crack down on distillation should be a top negotiation priority, but is hard to see why China would follow through, both because their companies rely on distillation to keep up, and because the practice is already inherently deceptive,” said Sam Hammond, director of artificial intelligence policy and chief economist at the Foundation for American Innovation, a center-right think tank.
Raising the issue could also complicate talks with China that the White House needs to go well.
Treasury Secretary Scott Bessent told CNBC on Monday that he and his Chinese counterpart had discussed “an incident line” for communications in the event of an AI safety breach. Bessent said they also plan on discussing protocols for both sides to agree on related to “the leading AI dangers,” including rogue agents and nonstate actors using bioweapons.
The administration will also be seeking economic concessions from the Chinese as voter anger over the economy continues to plague Republicans ahead of the upcoming midterm elections. In response to high tariffs imposed by the Trump administration, China restricted U.S. manufacturers’ access to rare earth metals critical for a range of domestic production. China agreed to lift those restrictions temporarily, and extending that pause is expected to be a top priority of administration officials.
Still, the push to do something about Chinese “distillation” has gained traction as evidence of the practice has grown. Anthropic released a report this month accusing Chinese firms Alibaba, Moonshot and DeepSeek of “illicit distillation.” Alibaba was accused of more than 151 million exchanges with Claude between May and July.
“Unauthorized distillation of American models by Chinese companies has very significant ramifications for the U.S.'s AI strategy — it is one of the most important reasons why Chinese models have remained only six to nine months behind their American counterparts,” said Nathan Calvin, general counsel at Encode, an AI advocacy organization. “It makes sense why this issue would be a top priority for the admin in these talks.”
The CISA report released Sept. 8, written with the FBI and National Security Agency, reflects the most extensive government documentation of the practice. It found that several Chinese firms had targeted the domain functions, reasoning capabilities and optimization techniques of Claude, ChatGPD, Gemini and Grok, “likely with Chinese government awareness.”
Even the Chinese models themselves appear confused. Earlier this year, one responded to a prompt by saying, “Hello! I’m Claude, an AI assistant made by Anthropic.”
Authors
