Trump official accuses Moonshot AI of distilling Anthropic’s Fable to train Kimi K3
White House science and technology chief Michael Kratsios has accused China’s Moonshot AI of using covert, large-scale distillation against Anthropic’s Fable model to help train its upcoming Kimi K3 system. The accusation came as Moonshot prepared to publish the weights of K3, a model released last week with benchmark scores at or near the U.S. frontier.
What Kratsios alleged
Kratsios said Moonshot built a sophisticated internal platform to run large-scale distillation against U.S. models. According to his account, the platform allowed Moonshot to quickly switch between multiple methods of access to avoid detection while siphoning capabilities from those models, including Anthropic’s Fable 5.
He also said Moonshot obtained advanced Nvidia chips, including GB300-equipped servers, and that the company is believed to have accessed GB300 hardware in Thailand, likely to train its AI models.
Position from the White House
In a public statement, the administration framed the issue as a line between legitimate and illegitimate distillation. The United States, the statement said, supports the free and fair development of AI, including a competitive ecosystem spanning frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate distillation used to create smaller, more efficient models was described as playing a vital role in that open innovation ecosystem.
Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research was labeled unacceptable.
Background with Anthropic
This is not the first time Anthropic has publicly named Moonshot. In February, Anthropic accused Moonshot, along with DeepSeek and MiniMax, of running industrial-scale campaigns to illicitly extract Claude’s capabilities. Sarah Heck of Anthropic’s policy team described distillation as IP theft and industrial espionage, tying the issue to national security risks.
Why K3 is at the center of the debate
Moonshot’s Kimi K3 was released last week with scores at or near the U.S. frontier. The full model is scheduled to have its weights published on July 27, which would place K3 among the most capable open-weight systems available from a Chinese AI lab. The combination of frontier-level performance and a forthcoming open-weight release is what makes the U.S. government’s allegation politically charged.
Why it matters
Confirmation of the underlying claims would shift the framing of K3’s release from a notable open-source milestone to a potential case of industrial espionage with geopolitical consequences. Critics have also noted that the accusation lines up with prior predictions that the U.S. would take action against leading open Chinese models, which is why verifiable evidence will be central to how the dispute is assessed.
FAQ
What did Michael Kratsios accuse Moonshot AI of doing?
Kratsios accused Moonshot AI of using a sophisticated internal platform to run large-scale distillation against U.S. models, including Anthropic’s Fable 5, while constantly switching access methods to avoid detection. He also said Moonshot obtained GB300-equipped Nvidia servers, including hardware accessed in Thailand.
How does this connect to Anthropic’s earlier complaint?
In February, Anthropic accused Moonshot, DeepSeek, and MiniMax of industrial-scale campaigns to illicitly extract Claude’s capabilities. Sarah Heck of Anthropic’s policy team described distillation as IP theft and industrial espionage linked to national security risks.
When is Moonshot releasing the K3 weights?
Moonshot released Kimi K3 last week with scores at or near the U.S. frontier. The full model is scheduled to have its weights published on July 27.