The White House has publicly accused Chinese AI startup Moonshot AI of covertly copying Anthropic's Fable model to build its newly released Kimi K3 system.

Michael Kratsios, director of the White House Office of Science and Technology Policy, said on July 22, 2026 that the US government has information that Moonshot AI "distilled Anthropic's Fable" during K3's development. He made the claim in a post on his official X account.

Distillation is a technique where a smaller model learns by training on the outputs of a larger, more capable one. Kratsios drew a sharp line between two versions of it. Legitimate distillation to build smaller, cheaper models, he said, is a normal part of open AI development. Large-scale covert extraction of proprietary US technology is not.

According to Kratsios, Moonshot built a sophisticated internal platform to run large-scale distillation against US models. He said the system was designed to rapidly switch between multiple access methods so the activity would evade detection.

He also alleged that Moonshot acquired servers equipped with Nvidia GB300 chips and accessed GB300 systems in Thailand, likely to train its models. That detail matters because advanced Nvidia chips face US export restrictions to China, and access routed through a third country raises questions about whether those controls were bypassed.

Kratsios did not disclose how the US government obtained the information or provide public evidence for the claims.

This is not the first time Moonshot has faced this accusation. Anthropic has previously alleged that Moonshot generated more than 3.4 million exchanges with its models through fraudulent accounts, describing it as a distillation attack built to bypass access limits and extract capabilities such as reasoning, coding, tool use, and computer vision at scale. Anthropic said metadata tied the activity to senior Moonshot employees.

Anthropic has also published guidance on detecting and preventing distillation attacks, part of a broader push by US frontier labs to make it harder for third parties to copy commercial models.

Kimi K3 is the model at the center of the dispute. Moonshot markets it as the first open 2.8 trillion parameter model, built for advanced reasoning, long-duration coding, and knowledge work, and priced at lower token costs than rival systems. Moonshot says K3 performs competitively with Anthropic's Fable 5 and beats several other leading models on its own benchmarks.

The allegations have not been independently verified. Kratsios offered no public proof, and Moonshot AI had not responded to the claims at the time of reporting. The accusation adds to rising tension over intellectual property in the US-China AI race, where the line between competition and theft is increasingly contested.

Sources: Michael Kratsios (@mkratsios47) statement on X, July 22, 2026; CyberScoop; France24; Anthropic. AI News MI will update this story if Moonshot AI responds.