White House accuses China’s Moonshot of stealing from Anthropic

TL;DR:

• White House science and technology adviser Michael Kratsios has accused China’s Moonshot AI of a “large scale” effort to copy top US AI models through distillation.
• He alleged Moonshot distilled capabilities from Anthropic’s Fable model to build its Kimi K3, and used restricted Nvidia servers to train it.
• Treasury Secretary Scott Bessent said sanctions “will be on the table” for industrial-scale distillation attacks.

The US has levelled a direct accusation of AI intellectual-property theft at a leading Chinese lab. Kratsios claimed Moonshot AI used distillation — extracting answers from a stronger model to train a weaker one — to lift capabilities from Anthropic’s Fable for its Kimi K3 model, which drew global attention last week for apparently narrowing the gap with Western systems.

Distillation as a flashpoint

The allegation folds two American anxieties together: model theft and chip smuggling. Kratsios said Moonshot likely trained K3 on Nvidia’s GB300 platform, whose most advanced chips have been export-restricted since 2022. Bessent had said a day earlier that the US would examine whether Chinese models stole from American rivals, drawing a line: “open source is not open season on American IP.”

The accusation is not isolated. In June, Anthropic accused Alibaba of “illicitly” extracting its Claude models’ capabilities through distillation attacks and urged Congress to penalise the firms involved. The White House said in April it would work more closely with US AI companies to counter “industrial-scale campaigns” to steal technology. Moonshot plans to release Kimi K3 as an open-source model on 27 July — the first of its scale to be freely downloadable.

Why the UK should watch

For UK readers, the dispute sharpens a dilemma rather than resolving it. Open-weight Chinese models are attractive precisely because they are cheap and unrestricted — appealing to British firms wary of over-reliance on US frontier labs. But an IP-theft cloud over their provenance, and the sovereignty friction already visible in US moves to control model access, complicate any adoption decision. The frontier race the accusation reflects is the same one UK safety evaluations keep probing.

Looking forward

China has not formally responded, and the claims rest on US government assertions yet to be substantiated publicly. With Trump due to meet Xi Jinping in September, the accusation raises the temperature on AI as a trade and security battleground — and on whether open-weight releases become the next front in it.