U.S. Treasury warns of sanctions as White House accuses Chinese AI firm of model theft

The U.S. Treasury is escalating its enforcement stance against Chinese artificial intelligence companies, with Secretary Scott Bessent declaring that sanctions remain a viable enforcement tool following accusations that Moonshot inappropriately extracted value from a leading American AI model.

The Accusation#

White House science and technology advisor Michael Kratsios alleged this week that the Beijing-based Moonshot engaged in systematic model distillation targeting Anthropic's Fable model. He also claimed the company had procured high-end Nvidia GB300 servers—chips subject to strict U.S. export restrictions—and potentially accessed additional units in Thailand to power its own AI development efforts.

What is Model Distillation?#

Model distillation represents a standard practice in AI development where smaller systems absorb knowledge by learning from a larger model's responses. While this technique has legitimate uses for efficiency and optimization, it can cross into intellectual property violation when applied without authorization at scale.

Treasury's Warning#

Bessent underscored his position in a post on X, stating that "Open source is not open season on American IP." He specifically warned that when Chinese firms "conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table."

Dispute Over Methods#

Some AI researchers have questioned whether Moonshot's recent open-weight model, Kimi K3, could have been built primarily through distillation from Fable given that Fable only became publicly available on July 1—just before K3's release. The advanced performance of Moonshot's model has prompted broader questions about whether U.S. companies can sustain their competitive lead and justify the massive investments fueling the AI industry's rapid expansion.

Broader Policy Debate#

The incident has reignited Washington's larger conversation about Chinese open-source AI models. Some officials, including Dean Ball—former White House AI advisor now at OpenAI—have advocated restricting or blocking Chinese open-weight models to preserve American technological competitiveness and address national security concerns.