1 hour ago
1217
A senior White House technology official on Wednesday accused Chinese startup Moonshot AI of covertly copying Anthropic’s Fable model to build its new Kimi K3 AI model, escalating a US-China conflict over AI in general just as Moonshot prepares to release the model to the public.
“Undermining American research is unacceptable”
Michael Kratsios, who directs the White House Office of Science and Technology Policy, laid out the claim in a post on X. He stated that his office had information that Moonshot ran its copying through a purpose-built internal system, while rotating access routes repeatedly to stay hidden as it pulled from US models.
Kratsios also said that the operation leaned on NVIDIA GB300 servers stationed in Thailand, which are Nvidia chips ranking among the chipmaker’s most powerful, and that US export rules bar from reaching China.
“Large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research is unacceptable,” Kratsios wrote in the post.
We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.
To do this they developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of…
— Director Michael Kratsios (@mkratsios47) July 22, 2026
Distillation: legal until it isn’t
The technique alleged to have been employed by Moonshot in the development of Kimi K3 is known as distillation. Distillation involves feeding the outputs of a stronger model to a weaker one, while training the weaker model to imitate the stronger one.
If done publicly and at a small scale, the practice is legal and common across the industry. However, the covert, industrial scale is widely objected to, and this is what Kratsios has accused Moonshot of doing. Kratsios says this is basically theft, and not routine engineering.
Wednesday’s accusations build on findings Anthropic published earlier in the year. In a February report, the company said it tracked more than 3.4 million Claude conversations back to Moonshot, routed through hundreds of fabricated accounts. Some of the account records lined up with the public profiles of Moonshot’s own senior staff, Anthropic stated.
The campaign, according to the company, went after Claude’s strengths in reasoning, coding, and vision.
Moonshot Kimi K3 model going public July 27
Moonshot put Kimi K3 out on July 16. At 2.8 trillion parameters, it is the largest open-weight model released to date, and it got so much interest that the company suspended new Kimi sign-ups within only two days.
The full model weights are due to go public by July 27, and Anthropic has said that this happening is a risk. The U.S. AI company argues that models built by copying drop the safety guardrails of the original model, and that once weights become open, they can travel beyond anyone’s control.
In January 2025, AI Czar David Sacks said DeepSeek had copied OpenAI’s models, an allegation DeepSeek outrightly rejected. Beijing also dismissed Anthropic’s February findings as groundless. Moonshot has said nothing publicly and Kratsios did not name any possible penalties.
The dispute also comes as the two governments prepare for AI talks in September. The US still outspends China on AI by almost 23 to one, yet Chinese labs keep closing the gap between U.S and Chinese AI models.
Don’t just read crypto news. Understand it. Subscribe to our newsletter. It's free.
English (US)