White House Accuses Moonshot AI of Distilling Anthropic's Fable for Kimi K3

The U.S. White House accused China-based Moonshot AI on July 22 of using AI distillation techniques to extract technology from Anthropic's advanced Fable model to develop its newly released Kimi K3 large language model. Michael Kratsios, director of the White House Office of Science and Technology Policy, stated that Moonshot built a sophisticated internal platform to conduct large-scale covert distillation, using restricted Nvidia GB300 chips and accessing servers in Thailand to train its models. U.S. Treasury Secretary Scott Bessent separately warned that Chinese companies could face sanctions if they conduct "covert industrial-scale distillation attacks" on U.S. technology.

Distillation is a technique in which a less advanced AI model is trained on outputs from a more capable model. The U.S. officials distinguished between legitimate distillation—used to create smaller, more efficient models—and what they characterized as large-scale, covert industrial theft of proprietary technology. Kratsios stated: "Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem. However, large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable."

China's Foreign Ministry, through spokesman Lin Jian, rejected the accusation on July 23, stating that the country "has always opposed politicizing and instrumentalizing science, technology, and trade issues." Lin Jian emphasized that Chinese AI development reflects "independent innovation" and international cooperation, and called for all nations to promote open, inclusive, and beneficial AI development.

AI researchers and experts expressed significant skepticism about the technical feasibility of the allegations. Braden Hancock, a researcher at the Laude Institute, noted that Fable had only been publicly available since July 1, making it implausible to distill sufficient data, train a model, and release it within two weeks. Nathan Lambert, an AI researcher at the Allen Institute for AI, argued that distillation becomes less impactful as Chinese models approach the frontier, and that advanced reinforcement learning techniques—which are much more resource-intensive—would be required to achieve Kimi K3's demonstrated capabilities. He observed that querying a frontier lab's API for such large-scale training "would be insanely expensive" and potentially infeasible.

Anthropic had previously accused Moonshot, DeepSeek, and MiniMax of systematically distilling its models earlier in 2026, citing millions of exchanges from IP addresses and meta data indicating deliberate capability extraction. The current accusation follows broader U.S. concerns about Chinese AI companies accessing U.S. technology, occurring as President Trump prepares for a September summit with Chinese President Xi Jinping. Moonshot announced plans to release Kimi K3 as an open-source model on July 27, making it freely available for developers and the public to download and customize.