A senior White House adviser has accused Chinese artificial intelligence company Moonshot AI of a "large scale" effort to steal capabilities from leading U.S. AI models, including those developed by Anthropic. Michael Kratsios, a science and technology adviser under former President Donald Trump, stated that Moonshot AI allegedly employed a technique known as distillation, where a less sophisticated model extracts knowledge from a more advanced one, to develop its K3 model.
Kratsios also claimed that Moonshot AI gained access to restricted, cutting-edge Nvidia servers to train its models. The allegations were made on social media platform X. Kimi K3, Moonshot AI's recently unveiled model, has garnered international attention for its perceived ability to close the gap between Chinese and Western AI technologies.
These accusations come as the U.S. government increases its scrutiny of Chinese AI companies. Just a day prior, Treasury Secretary Scott Bessent indicated that the U.S. would investigate whether Chinese AI models have misappropriated capabilities from American competitors. Bessent told Fox Business that if overseas models are found to be stealing from U.S. companies, the administration has the authority to impose sanctions.
The heightened U.S. focus on Chinese AI development is occurring in the lead-up to a planned meeting between former President Trump and Chinese counterpart Xi Jinping in September. Moonshot AI, Anthropic, the White House, and Nvidia have been contacted for comment.