A White House adviser has accused China's Moonshot AI of a large-scale effort to steal capabilities from top US artificial intelligence (AI) models. Michael Kratsios, US President Donald Trump's Science and Technology adviser, stated Moonshot AI carried out this campaign through a process known as distillation. Distillation involves a weaker AI model extracting answers and knowledge from a stronger, more advanced one.
Mr. Kratsios also alleged that Moonshot AI gained access to restricted, cutting-edge Nvidia servers to train its AI models. These accusations were made in a social media post on Wednesday. The US government has information that Moonshot AI 'distilled' capabilities from Anthropic's Fable AI for its K3 model. Moonshot's K3 model recently garnered global attention, with many observers believing it had significantly narrowed the gap with Western AI technology.
This development fits into a broader narrative of escalating technological competition and intellectual property disputes between the United States and China. Washington restricted the export of Nvidia's most advanced chips in 2022. This was due to concerns these chips could be used by the Chinese military. Governments worldwide have since increased efforts to prevent the smuggling of these critical components.
Mr. Kratsios' allegations follow similar concerns voiced by Treasury Secretary Scott Bessent. On Tuesday, Mr. Bessent stated the US would investigate whether Chinese AI models have stolen capabilities from American rivals. He added on Wednesday that sanctions 'will be on the table' for Chinese companies that 'cross the line' into intellectual property theft. Mr. Bessent specifically cited 'industrial-scale distillation attacks' as a red line. He emphasized that while open-source AI is supported, it does not grant 'open season on American IP'.
The Chinese government has not yet publicly responded to these specific allegations. However, China has previously maintained that its AI development is a result of its own dedication, effort, and international cooperation. These increased US scrutinies of Chinese AI companies occur ahead of an expected meeting between President Trump and Chinese counterpart Xi Jinping in September. The timing suggests AI and technology transfer will be a significant point of discussion.
Moonshot AI plans to release its Kimi K3 model as an open-source model on July 27. This would make it the first model of its scale freely available for developers and the public to download and customize. This move could further complicate intellectual property claims if the underlying technology is indeed found to be stolen. In June, Anthropic also accused Chinese e-commerce giant Alibaba of 'illicitly' extracting capabilities from its Claude AI models using distillation attacks. Anthropic urged the US Congress to penalize companies involved in such attacks and strengthen measures to prevent US technology theft.
The White House announced in April it would collaborate more closely with US AI companies. This collaboration aims to combat 'industrial-scale campaigns' by foreign rivals to steal technology. Mr. Kratsios noted in a memo at that time these campaigns sought to 'systematically undermine American research and development and access proprietary information'. The implications of these accusations are significant for the global AI landscape, potentially leading to stricter export controls and increased trade tensions. Investors and policymakers will closely watch the outcomes of these investigations and any subsequent sanctions.