Chinese startup Moonshot AI has found itself at the center of a major international scandal. White House AI advisor Michael Kratsios directly accused the company of secretly copying Anthropic's Claude Fable 5 model. According to him, the stolen data formed the basis of Moonshot's new flagship model, Kimi K3.

This is not about simple borrowing of ideas, but about systematic, industrial-scale theft. Kratsios stated that Moonshot created a specialized platform for mass distillation of American AI models. The essence of the method is training a new model on the responses of a more advanced one. Although distillation itself is a common practice, its covert and industrial application for technology theft, according to the White House representative, is unacceptable.

Scale of the Attack and Hardware Trail

Anthropic previously recorded over 3.4 million dialogues with Claude that led back to Moonshot. The traffic came from hundreds of fake accounts, and the data on their owners matched the profiles of the Chinese startup's top executives. The attack targeted Claude's intellectual, software, and visual capabilities.

A particularly sensitive aspect of the situation is the hardware angle. Kratsios noted that Moonshot obtained NVIDIA GB300 servers and used them in Thailand. It is worth recalling that the export of these top-tier AI chips to China is strictly prohibited by U.S. law. Washington tightened restrictions in May, and in June, raids were conducted at Super Micro's office in Taiwan over alleged chip smuggling. This story becomes another link in the chain where U.S. technology leaks to China in circumvention of sanctions.

Kimi K3 as a Catalyst for Confrontation

Moonshot unveiled Kimi K3 on July 16. The model contains 2.8 trillion parameters — making it the largest open-weight model to date. Demand was so high that the company suspended new subscriptions within 48 hours.

Anthropic warns that copied models lose their protective mechanisms, and once published, they cannot be controlled — they can spread without restrictions. Earlier, in January, similar accusations of copying OpenAI's models were leveled against DeepSeek. Beijing at the time called Anthropic's findings unfounded.

Moonshot remains silent for now, and Washington has not announced sanctions. However, the situation is crystal clear: China continues to expand its influence in AI, apparently using U.S. technology to do so. The next move is up to the American administration.

Expert opinion: This incident is not just a corporate conflict. It is a demonstration that U.S. technological leadership in AI is under direct threat. Chinese companies, using distillation and circumventing export restrictions, are capable of creating competitive models with a colossal number of parameters. The market should prepare for tighter controls and new rounds of the technology war, which will inevitably affect the cost and availability of advanced AI solutions.