White House Accuses Moonshot AI of Distilling Anthropic’s Fable for Kimi K3

Michael Kratsios, who leads the White House Office of Science and Technology Policy as Donald Trump’s Assistant for Science and Technology, accused Chinese AI company Moonshot AI on Wednesday of covertly distilling Anthropic’s Fable model to build its Kimi K3 system. Treasury Secretary Scott Bessent backed the claim soon after and raised the prospect of sanctions, turning a technical dispute over AI training methods into a formal diplomatic flashpoint.

Kimi K3 is a 2.8-trillion-parameter open-weights model that Moonshot AI released on July 16. Its quality rattled investors almost immediately: shares in US AI and chip stocks sold off on the Nasdaq within days of launch as traders weighed whether a cheaper, competitive open model out of China could dent demand for premium US model access.

Kratsios Alleges a Purpose-Built Distillation Pipeline

In a post on X on Wednesday, Kratsios wrote, “We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.” He went further, alleging the company built dedicated infrastructure for the effort: Moonshot AI “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.”

Kratsios also said Moonshot AI trained its models on Nvidia GB300 servers, a chip the United States does not permit selling to China, obtained either through new purchases or by accessing GB300 capacity in Thailand, according to CyberScoop. Kratsios did not disclose how the administration gathered its evidence, and Moonshot AI did not respond to CyberScoop’s request for comment.

Kratsios drew a line between the technique itself and the conduct he says crossed it: “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 wrote. “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.”

Treasury Puts Sanctions on the Table

Bessent posted his own statement on X shortly after Kratsios, opening with, “We support open-source AI and the innovation it unlocks.” He then pivoted to a warning, according to The Register: “But open source is not open season on American IP. When [People’s Republic of China] 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.”

The Entity List is the Commerce Department’s export control mechanism, and a listing generally requires a license for nearly all exports of US-origin technology to that organization, a broader restriction than the chip-specific GB300 export limits Moonshot AI is already accused of skirting. The US has already imposed sanctions or export restrictions on most of China’s major tech companies, according to The Register. Adding Moonshot AI to the Entity List would mark a sharp escalation beyond today’s chip-specific limits.

The Second Distillation Accusation Against Moonshot AI This Year

This is not the first time Moonshot AI has faced a distillation accusation tied to Anthropic’s models. The Register reported that Anthropic itself accused Moonshot AI of distillation in February 2026, alongside Chinese peers DeepSeek and MiniMax. This week’s accusation, coming from the White House and Treasury directly rather than from Anthropic, lands as Congress has its own inquiry already underway. In April, Representatives Andrew Garbarino (R-NY), who chairs the House Homeland Security Committee, and John Moolenaar (R-MI), who chairs the Select Committee on China, announced a joint investigation into what they called “a pattern of conduct by [Chinese]-based AI laboratories involving the large-scale theft of proprietary capabilities from American frontier AI systems through adversarial distillation,” the redistribution of those capabilities as open-weight models available for global download, and their incorporation into products used daily by American developers and engineers, per CyberScoop.

Piyush Sharma, CEO of AI security firm Tuskira, told CyberScoop the stakes extend past intellectual property. “When a model has learned to reason through software weaknesses, security gaps, and attack paths, copying its behavior also copies that analytical capability,” he said, pointing to the scale of a separate distillation campaign Anthropic previously attributed to Alibaba, which it said ran 28.8 million interactions with Claude from 25,000 fraudulent accounts over six weeks.

Moonshot AI has not disputed Wednesday’s accusation publicly. On its own website, the company describes Kimi K3 this way: “While its overall performance still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol, Kimi K3 demonstrated frontier-level performance across our evaluation suite, consistently outperforming other tested models.” Beijing has historically denied US industrial espionage allegations, and neither Moonshot AI nor the Chinese government had issued a formal response to the distillation claim at the time of publication.