The Chinese AI startup Moonshot AI allegedly used Anthropic's Fable model in developing its Kimi K3, according to Michael Kratsios, director of the White House Office of Science and Technology Policy.

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

"They developed a sophisticated internal platform for large-scale distillation of U.S. models, enabling them to quickly switch between various access methods to avoid detection. Moonshot AI also acquired servers equipped with GB300 and gained access to GB300 in Thailand, likely for training their AI models," he stated.

Kratsios did not disclose the source of this information or provide technical evidence. He distinguished between legitimate distillation and what he termed industrial-scale extraction of capabilities from American models. He emphasized that while legitimate use of the method is crucial for an open innovation ecosystem, large-scale covert distillation aimed at acquiring proprietary technologies is unacceptable.

On July 21, U.S. Treasury Secretary Scott Bessent hinted at sanctions against foreign developers found guilty of stealing American intellectual property. Following Kratsios's statement, he clarified that sanctions and inclusion on export restriction lists are among the potential measures being considered.

We support open-source AI and the innovation it unlocks. But open source is not open season on American IP. When PRC firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.

— Treasury Secretary Scott Bessent (@SecScottBessent) July 22, 2026

A representative from the Chinese embassy in Washington, Liu Chang, dismissed the accusations as "pure slander." He asserted that China advocates for the protection of intellectual property and urged the U.S. to stop discrediting the achievements of the Chinese AI industry.

Anthropic released Claude Fable 5 on June 9. On June 12, the U.S. government imposed export restrictions, leading the company to suspend access to the model for all users as it could not quickly verify their citizenship.

The authorities lifted the restrictions on June 30, and from July 1, Anthropic restored global access to Fable 5.

Moonshot AI unveiled Kimi K3 on July 16, just 15 days after the relaunch of Fable 5. Notably, Anthropic's product was also available for three days following its initial release in June.

Prime Intellect researcher Eli Bakouch pointed out that the short time frame makes it difficult to explain Kimi K3's performance, even if Moonshot AI did engage in distillation.

there are only 15 days between fable 5 ban removal and kimi K3 release. i don't think claiming that K3's performance comes from fable distillation (even if they did it) makes sense technically https://t.co/RQB8h07Qxo

— elie (@eliebakouch) July 22, 2026

Dean Ball, head of Strategic Futures at OpenAI, also stated that Kimi K3's quality cannot be solely attributed to distillation. Meanwhile, a screenshot circulated on social media showing the Chinese model allegedly introducing itself as Claude.

— Denise Wu (@denisewu) July 17, 2026

The authenticity and conditions under which the response was obtained have not been independently verified. The system instructions, previous messages in the dialogue, and model settings remain unknown.

Co-founder and CEO of Hugging Face, Clément Delangue, linked Kimi K3's success to the open model development. He believes that sharing results accelerates progress.

In February, Anthropic accused DeepSeek, Moonshot AI, and MiniMax of extracting capabilities from Claude through fake accounts. The developers claimed to have linked the campaigns to specific labs based on IP addresses, query metadata, infrastructure indicators, and, in some cases, data from partners. DeepSeek generated over 150,000 interactions, Moonshot AI over 3.4 million, and MiniMax around 13 million.

In June, reports emerged about renewed discussions within the Trump administration regarding restrictions on Chinese AI models with open weights. That same month, Anthropic raised separate concerns against operators linked to Alibaba and the Qwen lab.

Notably, in July, Tarun Chhabra, head of national security policy at a U.S. company, accused one of the leading Chinese AI tech firms, Zhipu, of training its GLM-5.2 model on responses from Claude and GPT.