U.S. Treasury Secretary Scott Bessent reiterated his stance on Wednesday, stating that sanctions remain a possibility against Chinese AI firms following a White House claim that Moonshot improperly distilled Anthropic’s Fable model. Model distillation, a common AI training technique, involves a smaller model learning from the outputs of a larger one. Bessent emphasized that open source does not equate to open season on American intellectual property, warning that covert, industrial-scale distillation efforts that cross into IP theft could face sanctions and Entity List designations. The Treasury official’s remarks follow a recent accusation from the White House’s science and technology policy chief, Michael Kratsios, who alleged that Moonshot engaged in large-scale distillation against U.S. models. Kratsios also claimed that Moonshot had accessed Nvidia’s GB300 servers in Thailand, which are part of the Blackwell generation and banned from sale to Chinese entities. Some experts question whether Moonshot’s K3 model, released last week as an open-weight model, was primarily developed through distillation from Fable, which became publicly available only in July. Moonshot’s advanced K3 model has raised concerns about the sustainability of U.S. AI labs’ business models amid the frontier AI race. The episode has intensified debates in Washington over the impact of Chinese open models on U.S. technological leadership and national security. TechCrunch has contacted Moonshot and the Treasury for comment.
The White House’s accusation against Moonshot centers on its alleged use of distillation techniques to access U.S. models, with specific mention of Nvidia’s GB300 servers. These servers, part of the Blackwell generation, are prohibited from being sold to Chinese companies. Kratsios alleged that Moonshot acquired these servers and used them in Thailand to train its AI models, potentially violating U.S. export control rules. The GB300 servers are a key component in high-performance AI training, and their use by Moonshot could signal a significant breach of U.S. regulations. However, some experts dispute the claim that Moonshot’s K3 model was primarily developed through distillation from Fable, noting that the model has only been publicly available since July 1. Moonshot’s release of K3 as an open-weight model has sparked broader discussions about the role of open models in the AI landscape and their implications for U.S. competitiveness.
The controversy highlights growing concerns in Washington about the influence of Chinese open models on the global AI race. Some policymakers, including former White House AI advisor Dean Ball, have argued that the U.S. should restrict or ban the use of Chinese open-weight models to protect its technological edge and address national security risks. The debate reflects broader tensions over intellectual property, export controls, and the ethical use of AI technologies in an increasingly competitive global market. The episode has also raised questions about the transparency and accountability of AI development processes, particularly when they involve cross-border collaborations and access to sensitive hardware.
Source: techcrunch
