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

BBC  |  Osmond Chia

US President Donald Trump's Science and Technology adviser Michael Kratsios has accused Beijing-based Moonshot AI of conducting a large-scale campaign to steal capabilities from Anthropic's Fable AI model. This alleged intellectual property theft was executed via distillation, a process where a weaker artificial intelligence model extracts advanced capabilities from a stronger competitor.

The controversy emerges just after the Kimi K3 model was unveiled last week, which reportedly narrowed the technological gap with Western counterparts. To train this system, the Chinese firm also allegedly bypassed export controls to access restricted Nvidia GB300 Grace Blackwell computing platforms. In response, US Treasury Secretary Scott Bessent warned that Washington is considering targeted sanctions against Chinese entities participating in these industrial-scale distillation attacks. This escalating technological confrontation occurs ahead of a planned September meeting between Trump and Chinese President Xi Jinping, even as the company prepares to release Kimi K3 as an open-source model on 27 July.

Comment

Export controls targeting physical hardware like the Nvidia GB300 Grace Blackwell platform face severe limitations against software-based knowledge transfer. By utilising distillation to train Kimi K3, developers capture the refined outputs of advanced Western models. Algorithmic extraction bypasses the need for massive supercomputing clusters. Consequently, Chinese AI capabilities can advance rapidly even while domestic fabrication plants struggle to replicate advanced silicon.

To read in details click here