US accuses Chinese AI startup Moonshot of stealing Anthropic technology

The administration of US President Donald Trump has accused the Chinese startup Moonshot AI of using technology from the American company Anthropic to create its new artificial intelligence model, Kimi K3.

This was reported by Reuters.

Washington also said it is considering imposing sanctions on the company.

What the US is accusing Moonshot AI of

Michael Kratsios, Director of the White House Office of Science and Technology Policy, said that Moonshot AI allegedly used a “distillation” method on Anthropic’s Fable model to develop Kimi K3. According to him, the company carried out a “large-scale covert copying” of American technology.

“We have information that Moonshot AI used Anthropic’s Fable to develop its K3 model. Large-scale covert distillation aimed at stealing American technology is unacceptable,” — Kratsios wrote on X.

Treasury Secretary Scott Bessent separately said he is considering adding Moonshot AI to the trade blacklist and imposing sanctions.

In response, Liu Chan, a spokesperson for the Chinese embassy in Washington, called the accusations unfounded. He said China respects intellectual property rights and urged American officials to “respect the facts and stop discrediting China’s achievements in artificial intelligence.”

Anthropic also backed the US authorities’ position. Company representative Sara Heck said that the unlawful use of American models by Chinese companies creates serious risks to US national security.

Kratsios also claims that Moonshot AI obtained servers with advanced Nvidia GB300 chips and used them in Thailand to train its models. This could indicate a possible violation of US export controls.

Moonshot AI has not yet officially commented on the allegations.

More about Kimi K3 from Moonshot AI

Kimi K3 — an artificial intelligence model presented by the Chinese company Moonshot on July 17, 2026.

According to the developers, it is the largest open AI model in the world with 2.8 trillion parameters. It is primarily oriented toward programming, complex agent tasks, and handling large volumes of information.

Kimi K3 performs tasks such as:

• GPU code optimization;
• creation of the MiniTriton compiler;
• development of a browser-based 3D game;
• design of a specialized AI chip;
• automatic reproduction of scientific research with the writing of thousands of lines of code;
• creation of interactive analytical reports, presentations, and dashboards;
• video editing and creation of animated explainer clips.

Such use cases require significantly more computational resources than ordinary chatbots.

On July 20, Moonshot AI temporarily suspended the creation of new paid subscriptions for its new AI, Kimi K3.

More about Fable 5 and Mythos 5 from Anthropic

On June 9, 2026, Anthropic presented and opened user access to its AI models Claude Fable 5 and Claude Mythos 5.

According to Anthropic, Claude Fable 5 is the company’s most powerful model among those available for general use.

The model showed strong results in programming, data work, document analysis, and scientific research.

However, Fable 5 will not fulfill certain requests in cybersecurity, biology, chemistry, and model distillation.

If the classifier detects a potential threat, the request will automatically be redirected to the less powerful Claude Opus 4.8 model.

The Mythos model is positioned as one of the most “advanced” in programming and “agent” AI — that is, AI capable of independently performing complex tasks.

It can find vulnerabilities in cybersecurity and potentially create or improve cyberattacks.

The Mythos model is also used by the US National Security Agency (NSA), despite the Pentagon earlier including the company on a list of risky suppliers.

On June 13, 2026, Anthropic turned off Fable 5 and Mythos 5. The reason was a US government export control directive requiring the closure of access to the products for all foreign nationals, regardless of whether they are inside the US or abroad, including the company’s foreign employees.

However, on July 1, the company restored access to these models: the US government lifted export controls on Fable 5 and Mythos 5.

Also read: OpenAI models accidentally broke Hugging Face’s system. How did the incident happen?