This week, another scandal erupted in the world of artificial intelligence. White House science and technology advisor Michael Kratsios publicly accused Chinese startup Moonshot AI of secretly copying Anthropic's Claude Fable 5 model. According to him, the material obtained this way became the foundation for developing their new model, Kimi K3. This statement, made Wednesday on social media platform X, adds fuel to the fire of an already tense technological rivalry between the US and China.

The Core of the Accusations: Industrial Distillation

At the center of the conflict is the technology of distillation, where a new AI model is trained on the responses of a more advanced one. Open use of distillation is common practice. However, according to Kratsios, Moonshot's actions go far beyond what is permissible. He claims the company created a special platform for mass distillation of American models, constantly changing access methods to avoid detection.

"We have information that Moonshot AI used distillation of Anthropic's Fable in developing its K3... However, mass, covert, and industrial distillation aimed at stealing American technology and undermining US research initiatives is unacceptable," stated Michael Kratsios.

Anthropic had previously detected this attack. In a February report, the company tracked 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 public profiles of top managers at the Chinese startup. The attack targeted Claude's intellectual, software, and visual capabilities.

Notably, Kratsios also pointed out that Moonshot obtained servers with top-tier NVIDIA GB300 AI chips and used them in Thailand. It is worth recalling that the export of such chips to China is strictly prohibited by US law, and Washington tightened these restrictions in May. In June, a raid was conducted at Super Micro's office in Taiwan related to alleged chip smuggling to China.

Kimi K3 and Escalation of the Rivalry

Moonshot introduced the Kimi K3 model on July 16. It contains 2.8 trillion parameters, making it the largest open-weight model to date. Demand was so high that within 48 hours, the company had to suspend issuing new subscriptions. The full model weights are promised to be released by July 27.

This date is receiving particular attention. Anthropic warns that copied models lose their protective mechanisms. Once published, they can no longer be controlled and may spread without restrictions.

Similar accusations have been made before. In January 2025, White House AI advisor David Sacks accused DeepSeek of copying OpenAI's models. The company denied this, and Beijing authorities called Anthropic's February findings baseless.

Kratsios has not yet announced any penalties, and Moonshot has not commented on the situation. Although the US still outspends China on AI by a factor of 23, Chinese models continue to gain influence. The next move now lies with Washington.

Expert Opinion: This incident is not just another scandal. It demonstrates a systematic approach by Chinese companies to bypass technological restrictions. The successful distillation of Claude and the use of banned NVIDIA chips indicate that China is not just catching up but actively seeking loopholes to accelerate its progress. If the US does not tighten controls, we may see "stolen" technologies form the basis for the next generation of Chinese AI giants.