- White House OSTP Director Michael Kratsios accused Moonshot AI of distilling Anthropic's Fable to build its 2.8-trillion-parameter Kimi K3 model [1] - Treasury Secretary Scott Bessent threatened sanctions and Entity List designations, stating 'Open source is not open season on American IP' [2] - Kratsios also alleged Moonshot acquired Nvidia GB300 servers and accessed similar hardware in Thailand, potentially violating U.S. export controls [1] - Anthropic documented 3.4 million fraudulent Claude exchanges traced to Moonshot in a February 2026 report, part of 16 million total exchanges across three Chinese labs [5] - Some AI researchers question whether K3 could have been built primarily through distillation of Fable, which only became publicly available on July 1
[3] The U.S. Treasury Department on Wednesday threatened to impose sanctions on Chinese AI startup Moonshot AI after the White House accused the company of distilling Anthropic's Fable model to develop its recently released Kimi K3 system. Treasury Secretary Scott Bessent said sanctions and Commerce Department Entity List designations remain possible when Chinese firms engage in what he called 'covert, industrial-scale distillation attacks' against American AI models [2].
The accusations came from Michael Kratsios, director of the White House Office of Science and Technology Policy, who wrote on X that the administration had information indicating Moonshot built 'a sophisticated internal platform to conduct large-scale distillation against U.S. models' while rapidly switching between access methods to avoid detection [1]. Kratsios separately alleged that Moonshot acquired Nvidia GB300-equipped servers and accessed similar hardware in Thailand — hardware that falls under the Blackwell generation banned for sale to Chinese companies under U.S. export controls
.
[1]The escalation marks the first time a senior U.S. official has explicitly linked a specific Chinese AI model to alleged intellectual property theft from a named American company and threatened economic penalties in response. Moonshot has not publicly responded to the accusations. Beijing has previously dismissed broader U.S. allegations of illicit distillation as baseless [3].
The Accusations #
Kratsios's statement pointed to Moonshot's Kimi K3 — a 2.8-trillion-parameter open-weight model released last week — as the product of unauthorized distillation from Anthropic's Fable. K3 has drawn attention for performing competitively with Fable 5 and outperforming several leading U.S. systems on selected benchmarks, with independent evaluations ranking it among the strongest available for web development [3].
The allegations build on a February 2026 report from Anthropic documenting industrial-scale distillation campaigns by three Chinese labs: DeepSeek, Moonshot AI, and MiniMax. In that report, Anthropic said Moonshot generated over 3.4 million fraudulent Claude conversations targeting agentic reasoning, tool use, coding, data analysis, and computer vision capabilities. The three firms collectively produced more than 16 million exchanges across approximately 24,000 fraudulent accounts [5].
Anthropic traced metadata from the campaigns to senior Moonshot employees and characterized the activity as 'a distillation attack designed to bypass access restrictions.' Independent researchers have also identified statistical similarities between K3 and Claude models, with one screenshot reportedly showing K3 identifying itself as 'Claude, an AI assistant made by Anthropic' [4].
The Sanctions Threat #
Bessent's response elevated the matter from an intellectual property dispute to a potential national security enforcement action. 'Open source is not open season on American IP,' Bessent posted on X. 'When firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table' [2].
Kratsios drew a distinction between legitimate and illegitimate uses of distillation, a standard machine learning technique in which a smaller model learns from the outputs of a larger one. 'Legitimate AI distillation used to create smaller, more efficient models plays a vital role in this open innovation ecosystem,' he said, but characterized large-scale covert industrial distillation as 'unacceptable' [4].
The administration is reportedly considering adding Chinese AI laboratories to the Commerce Department's Entity List — which would restrict their access to U.S. technology — and potentially banning Chinese AI models from American use [4]. No formal penalties have been announced.
The Hardware Dimension #
Beyond the distillation allegations, Kratsios's statement raised a parallel export-control concern. He alleged Moonshot acquired Nvidia GB300-equipped servers and accessed similar systems in Thailand, potentially violating U.S. rules that prohibit the sale of Nvidia's Blackwell-generation chips to Chinese entities [1].
Nvidia's GB300 servers represent the company's latest and most powerful AI training infrastructure. The chips are subject to U.S. export restrictions aimed at preventing China from accessing advanced AI hardware. If confirmed, the alleged acquisition would represent a separate violation from the distillation claims [1].
Expert Skepticism #
Some AI researchers have questioned the timeline underlying the administration's claims. Anthropic made Fable publicly available on July 1, 2026, and Kimi K3 appeared weeks later — a compressed window that skeptics say is insufficient for large-scale distillation to have been the primary development method for a 2.8-trillion-parameter model [2].
However, the administration's framing suggests Moonshot's distillation campaign predated Fable's public release and targeted earlier Claude models as well, consistent with Anthropic's February report documenting months of sustained extraction activity [5].
Anthropic's head of public policy, Sarah Heck, appeared to endorse the White House's action, quoting Kratsios's statement on X and thanking him for 'speaking out on the issue' [6].
What's Next #
The episode has intensified a broader debate in Washington over whether Chinese open-weight AI models pose a competitive and security threat to American technology firms. K3's release — and its strong benchmark performance — has raised questions about the sustainability of major U.S. AI labs' business models if comparable open-weight alternatives are freely available [2].
For now, the Treasury's threat remains rhetorical. No sanctions, Entity List additions, or formal enforcement actions have been announced. Moonshot AI has remained silent, and the Chinese government has not issued a specific response to the latest accusations [3]. The coming weeks will determine whether the administration follows through with concrete measures or uses the threat as leverage in broader U.S.-China technology negotiations.
Companies mentioned #
Further sources #
[1] TechCrunch — Treasury threatens sanctions after White House claims Moonshot dis… ↗
[[2] Yahoo Finance — Treasury threatens sanctions after White House claims Moonshot … ↗](https://finance.yahoo.com/technology/ai/articles/treasury-threatens-sanctions-white-house-204903502.html)
[[3] Crypto Briefing — White House accuses Moonshot AI of using Anthropic's Fable to… ↗](https://cryptobriefing.com/moonshot-ai-distillation-allegations/)
[[4] Glitchwire — Trump Administration Accuses Moonshot AI of Distilling Anthropic's… ↗](https://glitchwire.com/news/trump-administration-accuses-moonshot-ai-of-distilling-anthropics-fable-escalati/)
[[5] Anthropic — Detecting and preventing distillation attacks ↗](https://www.anthropic.com/news/detecting-and-preventing-distillation-attacks)
[6] Seeking Alpha — Kratsios says Moonshot built Kimi K3 through industrial distill… ↗
The stories that matter, in one email. Free — unsubscribe anytime.