The White House on Wednesday accused Beijing-based Moonshot AI of covertly distilling Anthropic’s Claude Fable model to train Kimi K3, and of running that training partly on Nvidia GB300 servers accessed through Thailand in violation of U.S. export controls. It’s the sharpest attribution the Trump administration has made in the model-theft arena to date, and it arrives roughly three weeks after Fable itself became publicly available on July 1.
Michael Kratsios, director of the White House Office of Science and Technology Policy, laid out the claim on X. “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 access to avoid detection.” In a separate post he said Moonshot “acquired GB300-equipped servers and has accessed GB300s in Thailand, likely to train its AI models.” The GB300, part of Nvidia’s Blackwell generation, isn’t permitted for sale to Chinese firms. Kratsios didn’t detail how the government knows.
Treasury Secretary Scott Bessent supplied the enforcement half. “Open source is not open season on American IP. When [Chinese] firms conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.” A day earlier, on Fox Business with Maria Bartiromo, Bessent had claimed “we are finding watermarks of our U.S. large language models on many of the Chinese models.”
Moonshot released Kimi K3 last week as an open-weight model, with full weights expected next week. On its own site the company describes K3 as the first open 2.8-trillion-parameter model, and concedes that “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.” Naming the American frontier models inside your own release notes is a curious flex when Washington is looking for evidence of derivation.
Some experts, TechCrunch reports, doubt a model of K3’s scale could’ve been trained primarily via distillation from Fable in a three-week window. Moonshot didn’t respond to requests for comment.
The alignment on display is itself the story. Anthropic policy head Sarah Heck thanked Kratsios on X, calling “illicit, adversarial distillation is IP theft and industrial espionage that supports adversary military and intelligence capabilities.” This is the same administration whose Pentagon moved in February to blacklist Anthropic, and whose export controls blocked Fable 5’s release. Anthropic first accused Moonshot of distillation attacks back in February, to little federal echo. The lab spent the spring cast as a policy adversary; on Wednesday it became the aggrieved party the state department of commerce wants to protect. TechCrunch notes K3’s release “called into question the underlying business models of leading U.S. AI labs,” which is the quieter reason the frontier-lab lobby and the export-control hawks are, for now, reading from the same script.
Sources
- https://www.bloomberg.com/news/articles/2026-07-22/white-house-official-says-moonshot-accessed-banned-nvidia-chips
- https://techcrunch.com/2026/07/22/treasury-threatens-sanctions-after-white-house-claims-moonshot-distilled-anthropics-fable/
- https://cyberscoop.com/white-house-accuses-moonshot-ai-anthropic-model-distillation/
- https://www.aol.com/articles/top-white-house-official-escalating-163443000.html
- https://thehill.com/policy/technology/5984510-white-house-moonshot-ai-anthropic-nvidia/