Crypto news

22.07.2026
15:37

Distillation Chinese-style: Moonshot AI accused of stealing Anthropic's technology for the Kimi K3 model

White House AI advisor Michael Kratsios has leveled serious accusations against Chinese startup Moonshot AI. According to him, the company secretly copied Anthropic's Fable 5 model and used the resulting developments to create its own Kimi K3 model.

This is not just a technical dispute — it is another round of technological confrontation between the US and China. Kratsios published a statement on platform X, stating that Moonshot developed a specialized platform for large-scale distillation of models from the US. The company allegedly constantly changed access routes to avoid detection.

White House blocks Moonshot AI distillation

At the center of the conflict is distillation: a technique that allows training a new AI on the responses of a strong model. Open application is considered legal and widespread. However, Kratsios believes that covert distillation on an industrial scale is theft.

"We have data that Moonshot AI conducted distillation of Anthropic's Fable to create its K3 model… However, large-scale covert distillation aimed at stealing American technology and undermining the US research base is unacceptable," said Michael Kratsios.

Anthropic holds the same position. According to the company's February report, specialists tracked over 3.4 million interactions with Claude linked to Moonshot. Traffic came from hundreds of fake accounts — the credentials matched the public profiles of senior Moonshot employees. The attack targeted Claude's reasoning, coding, and image processing capabilities.

Kratsios also stated that Moonshot purchased NVIDIA GB300 servers and used them in Thailand. These are top-tier NVIDIA chips for AI, whose export to China is prohibited by US regulations. Washington tightened control over AI chip exports in May. In June, a raid was conducted at the Super Micro office in Taiwan — authorities were checking whether chips were being smuggled to China.

Kimi K3 model intensifies US-China tech competition

Moonshot unveiled Kimi K3 on July 16. The model has 2.8 trillion parameters — the largest open-source AI ever released. Demand was so high that Moonshot suspended new Kimi subscriptions within 48 hours. All model weights are planned to be published by July 27.

The publication date matters: Anthropic warns that copied models lose their safety guardrails. Once released into the open, these versions can spread without control.

This is not the first such complaint. In January 2025, White House AI chief David Sacks accused DeepSeek of copying OpenAI models. DeepSeek denied this. Beijing called Anthropic's February findings unfounded.

Kratsios did not announce sanctions, and Moonshot is not commenting on the situation. The US advantage in AI investment remains enormous — spending is nearly 23 times higher than in China. Meanwhile, Chinese developments are rapidly closing the gap. The next move is up to Washington.

Analyst's opinion: This story is a classic example of how geopolitics infiltrates the technology sector. Distillation itself is not new, but when it becomes a tool to bypass export restrictions, the consequences could be far more serious than just legal battles. The AI market has long ceased to be solely about innovation — it is now an arena for strategic confrontation.