Chinese startup Moonshot AI has found itself at the center of a new geopolitical scandal in the field of artificial intelligence. According to data shared by Michael Kratsios, Director of the White House Office of Science and Technology Policy, the company used the Fable model from American Anthropic to develop its flagship model, Kimi K3.
The Mechanism of Hidden Distillation
Kratsios claims that Moonshot AI created a complex internal platform for large-scale distillation — a process in which a less powerful model is trained on the responses of a larger one. According to him, the platform allowed for rapid switching between access methods to avoid detection. Additionally, the startup purchased servers with GB300 chips and gained access to similar equipment in Thailand, likely for training its models.
The official drew a clear line between legitimate distillation, which is part of an open innovation ecosystem, and the industrial extraction of proprietary technologies. He called the latter unacceptable.
Threats of Sanctions and China's Response
Against the backdrop of these statements, U.S. Treasury Secretary Scott Bessent hinted at the possibility of imposing sanctions against foreign developers caught stealing American intellectual property. Possible measures under consideration include both sanctions and the inclusion of companies on export restriction lists.
Liu Chang, a representative of the Chinese Embassy in Washington, called the accusations "pure slander," emphasizing that China advocates for the protection of intellectual property, and called for an end to the discrediting of the achievements of the Chinese AI industry.
Timeline and Technical Doubts
The timeframe raises questions among experts. Anthropic released Claude Fable 5 on June 9, but after the introduction of export restrictions on June 12, access to the model was suspended. The restrictions were lifted on June 30, and global access was restored on July 1. Moonshot AI unveiled Kimi K3 on July 16 — just 15 days after access to Fable was restored.
Eli Bakush, a researcher at the startup Prime Intellect, noted that such a short time window makes it difficult to explain the performance of Kimi K3, even if distillation did occur. Dean Ball, head of Strategic Futures at OpenAI, also stated that the quality of the model cannot be explained by distillation alone.
A screenshot circulated on social media showing Kimi K3 allegedly introducing itself as Claude, but the authenticity and the conditions under which this response was obtained have not been independently confirmed. The system instructions and previous messages in the dialogue are unknown.
Hugging Face co-founder Clément Delangue attributed the success of Kimi K3 to an open development model, emphasizing that sharing results accelerates progress. It is worth recalling that as early as February, Anthropic accused DeepSeek, Moonshot AI, and MiniMax of extracting Claude's capabilities through fake accounts. At that time, Moonshot AI generated over 3.4 million interactions.
My analysis: This incident is part of a systematic U.S. campaign to protect technological sovereignty in the field of AI. However, the timeframe and lack of technical evidence make the accusations debatable. It is likely that we will witness a tightening of export controls and new sanctions, which will only accelerate the development of alternative supply chains and proprietary models in China.