US says Chinese Moonshot AI stole tech, eyes sanctions
A top White House adviser has accused China's Moonshot AI of stealing US technology to build its new Kimi K3 model, a revelation that threatens fresh sanctions just days before the software is released globally as open-source.
Michael Kratsios, US President Donald Trump's Science and Technology adviser, said on Wednesday that Moonshot AI ran a "large scale" campaign to extract capabilities from Anthropic's Fable AI. The Chinese company allegedly used this technique, known as distillation, to develop its Kimi K3 model. Kratsios also alleged that Moonshot gained improper access to restricted Nvidia servers powered by the GB300 Grace Blackwell computing platform to train its systems.
Moonshot is scheduled to release Kimi K3 as an open-source model on 27 July. This will make it the first model of its scale that can be freely downloaded and customised by developers and the public. For European businesses and software engineers, the launch had promised a powerful, cost-free tool to integrate into commercial products and compete with US tech giants.
The theft allegations now introduce severe compliance risks for any European firm that adopts the technology. Washington restricted the export of Nvidia's most advanced chips in 2022 over military concerns, triggering a global crackdown on smuggling. Utilising a model trained on sanctioned hardware or stolen IP could expose European companies to secondary US sanctions, disrupting their operations and frightening away investors.
These claims mark a sharp escalation in the global tech trade conflict. On Tuesday, Treasury Secretary Scott Bessent warned that sanctions "will be on the table" when Chinese companies cross the line into IP theft. "We support open-source AI and the innovation it unlocks," Bessent said. "But open source is not open season on American IP."
Moonshot maintains that its K3 model rivals top US technology. The Chinese government has previously stated its AI development is "the result of its own dedication and effort as well as international cooperation". The heightened scrutiny arrives ahead of an expected September meeting between Trump and Chinese President Xi Jinping, following a June warning from Anthropic that Alibaba used similar distillation attacks on its Claude models.
