Treasury Secretary warns of sanctions on Chinese AI models

US threatens action over Chinese AI development practices

Treasury Secretary Scott Bessent said Tuesday the Trump administration is examining whether Chinese companies built popular open-source AI models using stolen intellectual property from US competitors, signaling potential sanctions ahead.

Speaking on Fox Business, Bessent described the practice—known in AI circles as distillation—in blunt terms: "There's a very technical AI word for it called distillation, but you and I would call it theft."

The Treasury Secretary said the administration supports open-source AI development broadly but will not tolerate models built on intellectual property 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," Bessent said.

He added that US investigators have found evidence of copying: "We are finding watermarks of our US large language models on many Chinese models, and that's unacceptable."

The distillation controversy

Bessent's comments align with earlier accusations from Anthropic, which claimed Chinese companies including Moonshot AI (maker of Kimi K3), Minimax, and DeepSeek improperly copied its research. Anthropic alleged these firms accessed its Claude model "at scale while evading detection" and used that access for training that violated the company's terms of service.

The Treasury Secretary did not name specific companies in his remarks.

Moonshot AI's Kimi K3 model has disrupted the generative AI landscape since its release. Industry benchmarks show the Chinese model performs tasks like coding at or above the level of Anthropic's Fable 5 and OpenAI's GPT-5.6-Sol. Unlike these proprietary models, Kimi K3 is open-weight, meaning users can download and modify it. Moonshot has committed to releasing the model's full weights on July 27.

Why it matters

The potential for US sanctions on Chinese open-source AI models represents a significant escalation in technology competition between the two countries. Open-source models have emerged as a counterweight to the dominance of proprietary systems from Anthropic and OpenAI, offering developers and companies alternatives that can be customized and deployed without vendor lock-in. Sanctions could fragment the global AI ecosystem and force companies to choose between Chinese and Western AI technologies, similar to dynamics that emerged in semiconductors and telecommunications.

Industry divided on response

Bessent's stance has entered an active debate within the technology sector about how to respond to Chinese AI advances.

Dean Ball, a former senior AI advisor to President Trump who recently joined OpenAI as head of strategy, faced criticism for suggesting the White House might spread fear, uncertainty, and doubt to discourage adoption of Chinese open-source models. Prominent tech investor Bill Gurley countered in a Washington Post op-ed Monday that open-source models represent one of the few paths to preventing an AI duopoly controlled by Anthropic and OpenAI.

Anthropie itself has faced questions about its distillation complaints, given that its own models were trained on copyrighted materials that authors and media companies say were used without permission.

Bessent cited Meta, Nvidia, and Reflection AI as US companies pursuing open-source models, though Chinese firms currently lead the space. He emphasized that foreign models "have to be held to the same standard as US models are, but more importantly, are they using stolen materials?"

These details were first reported by Business Insider.