White House Accuses Moonshot of Distilling Fable and Using Banned Nvidia Chips

Michael Kratsios, director of the White House Office of Science and Technology Policy, accused Moonshot AI on Wednesday of distilling Anthropic’s Fable to develop Kimi K3, the 2.8-trillion-parameter model the Chinese company released in preview on July 16. He also alleged that Moonshot had acquired GB300-equipped servers and accessed GB300s in Thailand, likely to train its AI models. Kratsios offered no public evidence and did not explain how the government obtained the information.

Kratsios alleges Fable distillation and GB300 access

Kratsios wrote that Moonshot developed “a sophisticated internal platform to conduct large scale distillation against U.S. models,” allowing the company to switch among access methods to avoid detection. He distinguished “legitimate AI distillation” used to make smaller models from “large-scale, covert industrial distillation aimed at stealing proprietary U.S. technology and undermining American research,” which he called unacceptable.

Nvidia’s GB300 belongs to its Blackwell generation, and U.S. export controls bar sales of those processors to Chinese companies. Bloomberg described a May Bureau of Industry and Security memo affirming that the restriction applies to Chinese companies outside China. Officials are examining whether advanced processors moved to subsidiaries of Chinese firms through a possible loophole. Nvidia previously dismissed the idea of such a loophole and told Bloomberg it operates as though none exists; the company did not comment on Kratsios’s post.

Moonshot describes K3 as the largest open-weight model built to date. Arena’s Frontend Code leaderboard placed it first at 1,679 points, ahead of Fable 5. Fortune put K3’s price at $15 per million output tokens, compared with $50 for Fable.

Ball and Low differ on the role of distillation

Anthropic made Fable publicly available on July 1, according to TechCrunch, leaving 15 days before the K3 preview.

Dean Ball, a former White House AI adviser who now heads strategic futures at OpenAI, wrote that Kimi’s performance could not “be explained away by distillation or anything like that.” When challenged, he added that distillation “plays a role” but “clearly it’s not primary.” Hamish Low, a research associate at the Institute for AI Policy and Strategy, told The Hill it was “pretty overwhelming likely they are distilling from leading U.S. models.” Low pointed to K3 excelling at coding but not at cybersecurity, which he said is what he would expect from heavy distillation of coding data.

The Hill also reported that screenshots of purported Kimi 3 conversations circulated over the weekend, including one answer that identified the system as Claude. The outlet did not present the screenshots as verified evidence of how K3 was trained.

Moonshot had not responded to requests for comment from several news organizations by publication. The Chinese Embassy in Washington did not respond to Reuters.

Bessent puts sanctions and Entity List designations on the table

Treasury Secretary Scott Bessent posted Wednesday that “open source is not open season on American IP” and warned that sanctions or Entity List designations “will be on the table” when covert, industrial-scale distillation crosses into theft. A day earlier, he told Fox Business that officials were finding “watermarks” of U.S. models in Chinese systems and would examine them in the coming days or weeks.

Anthropic policy chief Sarah Heck thanked Kratsios and called illicit distillation “IP theft and industrial espionage that supports adversary military and intelligence capabilities.” Under Secretary of State Jacob Helberg wrote that the allegation was “more than a heist of invaluable American Intellectual Property.” In February, Anthropic accused DeepSeek, Moonshot and MiniMax of collectively generating more than 16 million Claude exchanges through about 24,000 fake accounts. It attributed more than 3.4 million exchanges to Moonshot and stated that request metadata matched public profiles of senior Moonshot staff. The February report concerned earlier access to Claude; it did not address Fable or K3.

Nvidia CEO Jensen Huang disputed broader calls to restrict Chinese open models in an Axios interview Tuesday, before Kratsios published his accusation. “Distillation, learning from AI, learning from other sources of knowledge, is fundamental to intelligence,” Huang stated. The outlet noted that Huang supported consequences for privacy or contract violations, but he called the Chinese models “excellent” and recommended their use. He added, “There’s no scenario where China runs U.S. companies off the road.”

Moonshot’s Hugging Face page lists July 27 as the expected date for the full K3 weights. Tom’s Hardware noted that its published performance figures remain Moonshot claims until the weights are available. The United States and China plan AI talks in September, with Bessent leading the U.S. delegation, according to Reuters.

China Considers Adding AI Model Weights and Chip Designs to Export List

China's Ministry of Commerce is consulting Alibaba, ByteDance, Zhipu and other domestic AI and chip companies on possible controls affecting model weights, key data used for training and chip designs,

The Implicator

Trump Officials Revive Push to Bar Chinese AI Models After Kimi K3

Four separate attempts to restrict Chinese AI models reached internal consideration inside the Trump administration last year and were killed before any took effect, Axios reported Monday, and parts o

The Implicator

Commerce Department Lifts Export Controls on Anthropic's Fable 5 and Mythos 5

Anthropic said Tuesday that the U.S. Department of Commerce has lifted the export controls it imposed on the company's Claude Fable 5 and Mythos 5 models on June 12, and that it would begin restoring

The Implicator

Marcus Schuler

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai