The US Treasury has threatened sanctions against Chinese AI company Moonshot AI after White House officials accused the firm of improperly distilling capabilities from Anthropic’s Fable model to build its own Kimi K3, escalating a policy debate over Chinese open-weight models that now cuts across multiple branches of the US government.
Treasury Secretary Scott Bessent posted on X that “open source is not open season on American IP,” warning that “when firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.”
Michael Kratsios, the White House’s science and technology policy chief, went further, accusing Moonshot of conducting “large-scale distillation against US models” and alleging the company acquired Nvidia GB300 servers, part of the Blackwell generation, which is banned for Chinese companies under US export controls, and accessed them in Thailand.
The accusations are serious. But they face a significant timing problem.
Anthropic’s Fable model has been publicly available only since July 1, 2026. Moonshot released Kimi K3 in mid-July, roughly two to three weeks later. AI researchers have expressed skepticism that large-scale model distillation, which typically requires extensive compute and iterative training cycles, could have been completed in that window as the primary development method for a frontier-capability model.
The episode has intensified a broader Washington debate over how to handle the influx of capable, low-cost Chinese open-weight models. Dean Ball, the former White House AI adviser who now heads strategic futures at OpenAI, has argued for restricting or banning Chinese open models to preserve US technological advantage. Other voices, including Nvidia CEO Jensen Huang and open-source AI lab Arcee, have argued that openness serves US interests better than containment.
For Moonshot, the stakes are existential. Entity List designation would cut off access to US technology and software, a potentially crippling blow for a company that relies on Nvidia hardware. The company has not publicly responded to the allegations.
The Treasury’s threat marks the first time the US government has explicitly linked model distillation to sanctions enforcement. If followed through, it would establish a precedent that training techniques common across the AI industry could trigger national security penalties when practiced by Chinese firms against American models.
Sources: Treasury threatens sanctions after White House claims Moonshot distilled Anthropic’s Fable (TechCrunch, July 22, 2026)