The White House accuses Moonshot AI of industrial distillation of Anthropic's Fable model.
Chinese AI startup Moonshot AI has found itself at the center of a new scandal involving intellectual property rights violations. White House Office of Science and Technology Policy Director Michael Kratsios stated that the company used Anthropic's Fable model to develop its flagship model, Kimi K3. This is not a simple case of borrowing, but a well-organized operation on an industrial scale.
According to Kratsios, Moonshot AI created a complex internal platform for large-scale distillation of American AI models. This infrastructure allowed for rapid switching between different access methods to avoid detection. Moreover, the company purchased servers with the latest GB300 chips and gained access to similar computing power in Thailand, which was likely used for model training. It is important to note that the official distinguished between legitimate distillation—an important tool for open innovation—and covert industrial extraction of proprietary technologies, which he deemed unacceptable.
Sanctions on the Table: U.S. Treasury Department's Response
The following day, U.S. Treasury Secretary Scott Bessent escalated the rhetoric, suggesting the possibility of sanctions and inclusion of companies on the Entity List if theft of American intellectual property is identified. "Open source is not an open season on American intellectual property," Bessent emphasized.
The Chinese Embassy in Washington reacted sharply, calling the accusations "pure slander." Spokesperson Liu Chang stated that China advocates for the protection of intellectual property and called for an end to discrediting the achievements of China's AI industry.
Timeline and Technical Doubts
Anthropic released Claude Fable 5 on June 9. After brief export restrictions imposed on June 12, the model became globally available again on July 1. Moonshot AI unveiled Kimi K3 on July 16—just 15 days after Fable access was restored. This compressed timeline has sparked skepticism among experts. Prime Intellect researcher Eli Bakush noted that even with distillation, such a short window hardly explains Kimi K3's high performance. OpenAI's Head of Strategic Futures Dean Ball supported this view, stating that the model's quality cannot be attributed solely to distillation.
A screenshot also surfaced on social media showing Kimi K3 allegedly introducing itself as Claude. However, the authenticity of this dialogue has not been independently confirmed, and the context and system instructions remain unknown. Hugging Face co-founder Clement Delangue, on the other hand, linked Kimi K3's success to an open development model, emphasizing that sharing results accelerates progress.
My analysis: This incident is not just another episode in the U.S.-China technological rivalry. It demonstrates that distillation has become a weapon in the geopolitical struggle for AI leadership. However, the technical arguments of skeptics deserve attention: 15 days is indeed extremely short for creating a competitive model, even with distillation. More likely, we are witnessing not so much direct copying as the use of open-source developments and Moonshot AI's own research, with the accusations being part of a political game. The market should prepare for further escalation and new regulatory barriers.