China's Moonshot AI stole from Anthropic, Trump tech adviser says - BBC

White House Science and Technology adviser Michael Kratsios accused China's Beijing-based Moonshot AI of a "large scale" effort to steal capabilities from top US AI models, in a post on X on Wednesday.
Kratsios alleged Moonshot used a technique called distillation—where a weaker model extracts answers from a stronger one—to "distill" capabilities from Anthropic's Fable AI while developing its Kimi K3 model.
He also claimed Moonshot likely accessed restricted Nvidia GB300 Grace Blackwell servers; Washington has curbed exports of Nvidia's most advanced chips since 2022 over concerns about Chinese military use.
Kimi K3 drew global attention after being unveiled last week, with observers suggesting it narrowed the gap between Western and Chinese AI, and Moonshot claiming it rivals leading US technology; the model is set for open-source release on 27 July.
China's foreign ministry spokesperson Lin Jian said the country opposes "politicising" trade and tech issues and urged a "people-centred approach," while the Chinese government had not directly responded to the latest allegations.
The accusations followed Treasury Secretary Scott Bessent's remarks that the US would examine whether Chinese models stole American capabilities, warning sanctions "will be on the table" for "industrial-scale distillation attacks," with the scrutiny coming ahead of an expected September meeting between Trump and Xi Jinping.