White House accuses China's Moonshot AI of stealing Anthropic's model

A senior White House official has accused Chinese AI firm Moonshot AI of illicitly copying the advanced artificial intelligence model of U.S. company Anthropic. Michael Kratsios, director of the White House Office of Science and Technology Policy, stated on X that the U.S. has "information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model."

“

We have information that Moonshot AI distilled Anthropic’s Fable for the development of its K3 model.

AI model "distillation" involves training a smaller AI to mimic the outputs of a larger, more capable one. U.S. officials allege that Chinese firms are using this method extensively to steal proprietary American systems. Kratsios further claimed Moonshot AI built a "sophisticated internal platform to conduct large-scale distillation against U.S. models," enabling the company to "quickly switch between multiple methods of access to avoid detection."

“

Moonshot built a sophisticated internal platform to conduct large-scale distillation against U.S. models, which allowed the company to quickly switch between multiple methods of access to avoid detection.

Adding to the concerns, Kratsios indicated that Moonshot AI acquired servers with Nvidia's advanced GB300 chips and accessed similar servers in Thailand. This suggests an attempt to bypass U.S. export controls that limit Chinese access to high-end American semiconductors. While Kratsios did not specify potential consequences, Treasury Secretary Scott Bessent recently stated that foreign companies could face sanctions for stealing U.S. technology. Bessent noted that "watermarks" of U.S. large language models have been found on Chinese models, deeming it "unacceptable."

“

If we see, especially that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft.