A senior White House official has accused China’s Moonshot AI of covertly stealing Anthropic’s artificial intelligence technology and accessing restricted Nvidia chips through Thailand-based servers to build its breakout Kimi K3 model.
In a July 22 post on X, Office of Science and Technology Policy Director Michael Kratsios said that his office has reason to believe that Moonshot AI “distilled Anthropic’s Fable for the development of its K3 model.”
Its arrival sent chip stocks tumbling in Asia and the United States. Days later, the company that built it could not keep the model running for new users.
Kratsios said that to do this, “Moonshot developed a sophisticated internal platform to conduct large scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection.”
IBM stated, “Knowledge distillation is a machine learning technique used to transfer the learning of a large pre-trained ‘teacher model’ to a smaller ‘student model.’”
Access to both models was suspended on June 12 after the government issued an export control directive barring their use by foreign nationals, including Anthropic employees who are noncitizens.
Kratsios said that Moonshot has also acquired servers equipped with Nvidia’s GB300 system and “has accessed GB300s in Thailand, likely to train its AI models.”
Washington has sought to restrict Beijing’s access to sales of Nvidia’s H200 chips that could be used for military applications.
Kratsios said that the United States “strongly supports the free and fair development of AI, including a thriving competitive ecosystem that spans frontier models, specialized systems, open-source frameworks, and open-weight models.”
He said that 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,” he said.
Moonshot released Kimi K3 on July 17 and showed it off at the World Artificial Intelligence Conference in Shanghai.
The company described it as the largest openly available AI model ever built, with 2.8 trillion parameters, the internal settings a model uses to weigh information. It could take in about a million tokens, the chunks of text a model reads, in a single request, enough to hold an entire software project at once.
Moonshot said the model performed on par with Anthropic’s Claude Fable 5, one of the most capable systems available, and outperformed OpenAI’s GPT-5.6 Sol and Anthropic’s Claude Opus 4.8.
It also costs far less. Moonshot charged about $3 for every million tokens fed into the model and $15 for every million it produced, compared with roughly $50 for the same volume of output from Fable 5.
Anthropic’s head of public policy, Sarah Heck, replied to Kratsios’s post thanking him “for speaking out on this important issue” in a July 23 post on X.
“Illicit, adversarial distillation is IP theft and industrial espionage that supports adversary military and intelligence capabilities,” she said.
“It is a national challenge that creates serious national security risks for the United States and democratic allies. We'll continue to crack down on it and work with the Administration and Congress to maintain American AI leadership.”
The Epoch Times contacted Moonshot AI and Nvidia for comment.
