The U.S. Treasury is threatening sanctions against Chinese AI companies, and the trigger is a direct accusation that Moonshot stole from Anthropic. According to TechCrunch AI, Treasury Secretary Scott Bessent doubled down on Wednesday, saying sanctions remain on the table after a White House official accused Moonshot of improperly distilling Anthropic’s Fable model. This is the sharpest escalation yet in Washington’s fight over Chinese open-weight AI.
What actually happened
Bessent’s warning followed a specific charge from Michael Kratsios, the White House’s science and technology policy chief. TechCrunch AI reports that Kratsios accused Moonshot of running “large-scale distillation” against U.S. models. He went further, alleging the firm acquired Nvidia’s “GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.”
That second claim matters as much as the first. GB300 servers are part of Nvidia’s Blackwell generation, which is banned from sale to Chinese companies. So there are two accusations stacked here:
- IP theft through distillation of Anthropic’s Fable
- Export-control evasion by allegedly sourcing banned chips through Thailand
Bessent’s line on X was blunt: “Open source is not open season on American IP.” He added that “covert, industrial-scale distillation attacks that cross the line into IP theft” would put “sanctions and Entity List designations” on the table.
What distillation is, and why the timing is suspicious
Distillation is a common, mostly legitimate technique. A smaller model learns from the outputs of a larger one to get cheaper and faster without training from scratch. It can cross into IP infringement, but labs use it every day as a normal optimization step.
Here’s what stands out. Some experts told TechCrunch AI they doubt Moonshot’s Kimi K3 could have been built primarily by distilling Fable, because Fable has only been public since July 1. Moonshot released K3 as an open-weight model last week. Building a frontier-class model off a few weeks of access to a rival is a hard story to tell. That gap between the accusation and the timeline is why this reads more like a policy fight than a settled case.
Why this matters for the industry
K3’s real threat isn’t just legal. Its capabilities have called into question the business models of leading U.S. AI labs. If an open-weight Chinese model can match frontier performance, it undercuts the argument that justifies the enormous capital U.S. labs are raising and spending. That’s the uncomfortable subtext under the IP language.
The episode has pushed a bigger debate into the open. Dean Ball, a former White House AI adviser now serving as OpenAI’s Head of Strategic Futures, has argued the U.S. should restrict or effectively ban Chinese open-weight models to protect its lead and reduce national security risk. That position was fringe a year ago. It isn’t anymore.
What to expect next
A few things are worth watching:
- Entity List action. If Moonshot lands on the Entity List, U.S. firms would need licenses to do business with it, and other Chinese labs would take notice.
- The chip angle. The Thailand allegation opens a separate export-control investigation that could reach Nvidia’s distribution partners, not just Moonshot.
- Open-model policy. Expect louder calls in Washington to curb the use of Chinese open-weight models inside U.S. companies and government.
For practitioners, the practical read is simple. If your stack leans on Chinese open-weight models like Kimi K3, watch this closely. A sanctions designation or a usage restriction would change what you can legally deploy, fast.
Moonshot and the Treasury had not commented when TechCrunch AI reached out. This one is still developing, and the response from Beijing and from Nvidia will shape where it goes. Full details are at the original TechCrunch AI report.