On July 22, 2026, US Treasury Secretary Scott Bessent said the federal government will scrutinize Chinese open-weight AI models for evidence of stolen intellectual property, with sanctions on their makers potentially following within weeks. The same day, Nvidia chief executive Jensen Huang publicly disagreed, calling Chinese models “excellent” and arguing that American firms should be free to run them.
What Bessent announced
In a television interview reported by Bloomberg, Bessent framed the policy as a defense of US innovation. “This administration supports open source models, but what we do not support is IP theft,” he said. His stated evidence is technical. “We are finding watermarks of our US large language models on many Chinese models,” Bessent said, referring to a technique called distillation, in which one model is trained on the outputs of another. Bessent called the practice theft and promised enforcement action in a matter of weeks.
What triggered the warning
The immediate cause was Kimi K3, an open-weight model from China’s Moonshot AI. According to the source reporting, Kimi K3 matches or beats the leading models from OpenAI and Anthropic on some benchmarks while costing significantly less. Its release contributed to a selloff in chip stocks and renewed concern in Washington that US dominance in frontier AI is narrowing. US laboratories have accused Chinese competitors of siphoning their models for months.
Huang’s rebuttal
Speaking to Axios at a new chip facility in Texas, Nvidia’s Jensen Huang dismissed the panic as misplaced. “These Chinese models are excellent,” he said. “Open-source models that are excellent should be used.” American firms, he argued, should “absolutely” be allowed to run them. Huang’s argument is commercial: cheaper or free models put AI into more hands, and broader use translates into more demand for the chips, data centers, and power generation that Nvidia sells. “Free AI should be great for chips,” he said.
Huang also pushed back on the idea that a downloaded Chinese model functions as a “backdoor” for Beijing, noting that firms can isolate such models inside secure sandboxes.
Why the dispute is awkward for US labs
Not every industry figure accepts the theft framing. Hugging Face chief executive Clem Delangue told TechCrunch that distillation is a minor and widely shared practice. “We know distillation to be a very small factor,” he said, adding that it is “a practice that everyone is doing, including companies in the US.”
There is a further complication. Anthropic, one of the most vocal accusers of Chinese distillation, recently had a $1.5 billion settlement approved over pirated books used to train Claude, the largest copyright settlement in US history. Anthropic was also named in a separate lawsuit by the University of Tennessee over neural network patents, and publisher Bloomsbury is among those in line for a payout from the books settlement. Huang extended the same permissive logic to Anthropic’s upcoming Mythos model, telling the government to “let Anthropic run.”
What happens next
Bessent will lead the US delegation at AI talks with China scheduled for September, where the distillation dispute is expected to be on the agenda. For now, the split is clear: the US government wants to fence Chinese models out, while the world’s most valuable chipmaker wants them let in.
FAQ
What did Treasury Secretary Scott Bessent say about Chinese AI models?
Bessent said the US will scrutinize Chinese open-weight models for stolen intellectual property, citing watermarks from US large language models found on Chinese systems. He framed distillation as theft and said sanctions on makers could follow within weeks.
Why is Jensen Huang defending Chinese AI models?
Huang, the chief executive of Nvidia, told Axios that Chinese models are “excellent” and that US firms should “absolutely” be allowed to run them. His reasoning is commercial: broader AI adoption increases demand for Nvidia’s chips, data centers, and power infrastructure.
What is distillation in AI, and is it really theft?
Distillation is the practice of training one model on the outputs of another. Bessent and some US labs describe it as intellectual property theft. Hugging Face’s Clem Delangue told TechCrunch that it is “a practice that everyone is doing, including companies in the US,” and called it a “very small factor” in competitive dynamics.
Related coverage
This article summarizes reporting from thenextweb.com. See our editorial disclaimer for how our articles are produced.
Run a free scan to see your AI Visibility Score, SEO rating, and local citation accuracy.