Treasury Warns Chinese AI Firms Could Face Sanctions Over Alleged Model Distillation

U.S. Treasury Secretary Scott Bessent said Wednesday that sanctions remain a possibility for Chinese artificial intelligence companies following allegations from the White House that Moonshot improperly distilled Anthropic’s Fable AI model. The comments add to growing scrutiny in Washington over the development of advanced AI models by Chinese firms and the protection of U.S. intellectual property.

Model distillation allows a smaller AI model to learn from the outputs of a larger one. While the technique is widely used to improve efficiency, it can also raise intellectual property concerns when conducted without authorisation.

Treasury warns of possible sanctions.

In a post on X, Bessent said open-source AI should not be treated as unrestricted access to American intellectual property. He stated that sanctions and Entity List designations remain possible if Chinese companies conduct what he described as large-scale distillation that crosses into IP theft.

Earlier this week, Bessent said the U.S. government would examine Chinese open-source AI models for evidence of intellectual property violations and consider sanctions if such activity is found.

White House raises concerns over Moonshot.

Bessent’s latest remarks followed comments from White House science and technology policy chief Michael Kratsios, who accused China-based Moonshot of carrying out large-scale distillation of U.S. AI models. Kratsios also alleged that the company obtained Nvidia GB300-equipped servers and accessed additional GB300 systems in Thailand, raising questions about possible violations of U.S. export controls.

Nvidia’s GB300 servers are part of the Blackwell generation of AI hardware, which is prohibited from being sold to Chinese companies under U.S. export restrictions.

Debate over Chinese AI models continues.

Some AI experts have questioned whetherMoonshot’ss recently released Kimi K3 model could have been developed primarily through distillation from Anthropic’s Fable model, noting that Fable has only been publicly available since July 1. Kimi K3 was released last week as an open-weight model.

The dispute has intensified debate in Washington over Chinese open-weight AI models. Some policymakers and industry figures, including former White House AI adviser and current OpenAI Head of Strategic Futures Dean Ball, have argued that the United States should restrict or effectively ban the use of Chinese open-weight models because of intellectual property and national security concerns.