Beijing rejected Washington’s accusations and vowed to protect the interests of Chinese AI companies.
China’s Ministry of Commerce has accused the United States of pursuing “AI hegemonism” and threatened countermeasures after U.S. officials said Chinese AI companies could face investigations, sanctions and trade restrictions over the alleged theft of American AI technology.
The ministry said the accusations are based on claims that Chinese developers used model distillation — an AI training technique in which the outputs of a more capable model are used to train or improve another system. Beijing argued that Washington lacks both factual and legal grounds to claim the technology was used to steal U.S. intellectual property.
The latest escalation centers on Beijing-based Moonshot AI, whose recently released Kimi K3 model has intensified debate in Washington over whether Chinese developers are copying U.S. AI models or rapidly closing the technological gap through their own research. The controversy echoes the U.S. reaction to the rise of Chinese startup DeepSeek last year.
White House Office of Science and Technology Policy Director Michael Kratsios said the U.S. government has information indicating that Moonshot distilled Anthropic’s Claude Fable 5 model to develop Kimi K3. He also alleged that the company built a sophisticated internal platform for large-scale distillation of U.S. models and used servers equipped with Nvidia GB300 chips, including systems located in Thailand, to train its AI models.
U.S. Treasury Secretary Scott Bessent later warned that Chinese developers could face financial sanctions or be placed on the Commerce Department’s Entity List, a move that would severely restrict their access to U.S. semiconductors, software and cloud services. “We support open-source AI and the innovation it unlocks. But open source is not open season on American IP,” Bessent said.
Moonshot rejected the allegations, saying Kimi K3’s performance was achieved through original changes to its underlying architecture rather than model distillation. Meanwhile, Anthropic said it had identified more than 3.4 million interactions with its Claude models linked to Moonshot, alleging that hundreds of fraudulent accounts were used to target capabilities including reasoning, coding, data analysis, tool use and computer vision.
Source: Reuters
