China’s Moonshot AI stole from Anthropic, Trump tech adviser says
A White House adviser has accused China’s Moonshot AI of a “large-scale” effort to steal the capabilities of the United States’ top artificial intelligence (AI) models.
Michael Kratsios, US President Donald Trump’s Science and Technology advisor, said Moonshot AI ran the campaign using what is known as distillation, where a weaker AI model extracts the answers of a stronger AI model.
Moonshot also gained access to limited state-of-the-art Nvidia servers to train its models. Kratsios said in a social post:, external on Wednesday.
The BBC has contacted Moonshot, the Chinese embassy in Washington, Anthropic, the White House and Nvidia for comment.
Kratsios said at X that the US government had information that Moonshot AI had “distilled” capabilities from Anthropic’s Fable AI for the development of the K3 model.
Kimi K3 gained worldwide attention after its introduction last week, with many believing it narrowed the gap between Western and Chinese AI models. Moonshot said its K3 model could rival the best US technology.
Moonshot will also likely have used servers powered by Nvidia’s GB300 Grace Blackwell computing platform, Kratsios said.
Washington restricted the export of Nvidia’s most advanced chips in 2022 due to concerns that they could be used by the Chinese army.
Since then, governments around the world have cracked down on chip smuggling.
Kratsios’ claims come a day after Treasury Secretary Scott Bessent said Tuesday that the United States would examine whether Chinese artificial intelligence models are stealing talent from American rivals.
On Wednesday, Bessent also said sanctions would be “on the table” if Chinese companies “cross the line” on intellectual property (IP) theft by carrying out “distillation attacks on an industrial scale.”
“We support open source AI and the innovation it unlocks. But open source is not open season on American intellectual property,” Bessent said on social media.
The Chinese government has yet to issue a press release in response to the latest allegations, but has previously said its country’s AI development is “the result of its own dedication and effort as well as international cooperation.”
The increased US scrutiny of Chinese AI companies also comes as Trump is expected to meet with his Chinese counterpart Xi Jinping in September.
Moonshot AI will release the Kimi K3 as an open source model on July 27, making it the first model of its scale that can be downloaded and customized for free by developers and the public.
In June, Anthropic accused Chinese e-commerce and tech giant Alibaba of “illegally” extracting the capabilities of Claude AI models using distillation attacks.
Anthropic called on Congress to punish the companies behind the alleged attacks and increase measures to prevent the theft of US technology.
The White House said in April it would work more closely with US artificial intelligence companies to combat “industrial-scale campaigns” by foreign rivals to steal the technology.
The campaigns aimed to “systematically undermine American research and development and access to proprietary information,” Kratsios said in a memo at the time.
