Key facts
- China's commerce ministry accused the U.S. of "AI hegemonism" and threatened countermeasures.
- U.S. officials allege Chinese AI companies may face investigations, sanctions, and trade restrictions.
- The dispute involves allegations that Moonshot AI's Kimi K3 model was developed using "distillation" of U.S. AI models.
- Moonshot AI denies the allegations, attributing Kimi K3's performance to original architectural changes.
- U.S. officials cited potential acquisition of Nvidia GB300 chips in Thailand for model training.
- Sanctions and Entity List designations are being considered by the U.S. for alleged IP theft.
China's commerce ministry has accused the United States of "AI hegemonism" and threatened countermeasures in response to potential U.S. investigations into Chinese AI companies for alleged technology theft. The ministry stated that Washington's threats against Chinese firms, based on allegations of using "distillation" to copy advanced U.S. AI models, lack factual or legal grounds.
Model distillation, a common AI training technique, is at the center of the dispute. U.S. officials, including White House Office of Science and Technology Policy Director Michael Kratsios, have alleged that Beijing-based Moonshot AI used distillation to develop its Kimi K3 model, potentially copying Anthropic's Claude Fable 5. Kratsios also suggested Moonshot acquired servers equipped with Nvidia's GB300 chips in Thailand for model training.
U.S. Treasury Secretary Scott Bessent warned that Chinese companies engaging in "covert, industrial-scale distillation attacks that cross the line into IP theft" could face financial sanctions or be placed on the Commerce Department's Entity List. Moonshot AI has rejected these claims, asserting that Kimi K3's performance improvements stem from original changes to its underlying architecture. Anthropic reported identifying over 3.4 million interactions with its Claude models linked to Moonshot, utilizing hundreds of fraudulent accounts.