Kimi K3 Under Fire in the USA: Was the Model Distilled from Claude Fable of Anthropic?

The clash between the United States and China on the front of artificial intelligence is intensifying. The American administration has indeed launched accusations against the Chinese startup Moonshot AI, claiming that the company has improperly developed the Kimi K3 model by leveraging distillation techniques applied to the Claude Fable model of Anthropic.

The issue revolves around the so-called "distillation" of artificial intelligence models, a well-established technique that involves training a smaller model using the responses generated by a more advanced one. The "student" model thus learns the behavior of the "teacher" model, achieving high performance with lower computational costs. This is a widely used practice in research and commercial development when it is done in compliance with the licenses and authorizations required.

According to Washington, however, Moonshot AI has crossed this boundary. Michael Kratsios, director of the Office of Science and Technology Policy at the White House and scientific adviser to President Donald Trump, publicly accused the Chinese company of having built an infrastructure dedicated to the systematic extraction of capabilities from U.S. models. In simple terms, the allegation is that the company appropriated American technology to develop its flagship model presented in recent days.

Kratsios stated: "We have information that Moonshot AI distilled the Fable model of Anthropic for the development of its K3 model. To do this, it developed a sophisticated internal platform capable of performing large-scale distillation operations against U.S. models, allowing rapid switching between different access methods to evade detection. Moonshot AI also acquired servers equipped with GB300 and had access to GB300 systems in Thailand, possibly used to train its AI models."

"The United States strongly supports free and fair development of artificial intelligence, including a competitive ecosystem that includes frontier models, specialized systems, open-source frameworks, and open-weight models. Legitimate AI distillation, used to create smaller and more efficient models, plays a crucial role in this open innovation ecosystem. However, industrial-scale distillation conducted covertly with the aim of stealing proprietary U.S. technology and undermining American research is unacceptable."

A few hours after the White House's statements, Scott Bessent reiterated the American administration's stance through a message posted on X, stating that "open source does not mean free availability of American intellectual property." The Secretary of the Treasury added that if Chinese companies carry out covert industrial distillation operations that constitute intellectual property theft, the United States will consider imposing sanctions and placing the involved companies on the Entity List. Just days prior, Bessent had indicated that the government would examine open-source models developed in China for potential violations of intellectual property.

However, not all experts share the view of the accusations. Some observers find it unlikely that Kimi K3 was predominantly developed through the distillation of Fable, considering that the Anthropic model has only been publicly available since July 1. Moonshot AI has distributed Kimi K3 as an open-weight model, and its performance has reignited the debate on the economic sustainability of large U.S. laboratories, questioning the necessity of the enormous investments required for the development of frontier models.

In recent months, the American administration had already reported alleged campaigns organized by Chinese actors aimed at extracting the knowledge embedded in models developed in the U.S. Anthropic had reported detecting millions of suspicious interactions related to Moonshot AI, DeepSeek, and MiniMax, while some tests conducted after the launch of Kimi K3 had shown responses in which the model identified itself as Claude. These elements have fueled suspicions but do not currently represent definitive proof of how the Chinese model was trained.

Finally, the affair fuels the ongoing debate in the United States about the appropriateness of limiting the use of open-weight models developed in China. Some industry figures, including Dean Ball, a former White House adviser on artificial intelligence and now Head of Strategic Futures at OpenAI, believe that stricter restrictions could help preserve the American technological edge and reduce potential risks to national security.