China’s Moonshot AI stole from Anthropic, Trump tech adviser says
A White House technology adviser has accused Chinese firm Moonshot AI of running a "large scale" effort to steal capabilities from leading US artificial intelligence models, intensifying scrutiny of Chinese AI developers amid the ongoing US-China tech rivalry. Michael Kratsios, Donald Trump's Science and Technology adviser, said Moonshot used "distillation" – a technique where a weaker AI model learns by extracting answers from a stronger one – and had also gained unauthorised access to restricted, cutting-edge Nvidia servers to train its systems.
Kratsios said the US government has evidence that Moonshot AI distilled capabilities from an Anthropic model to help build its Kimi K3 system, which was unveiled last week and drew global attention for apparently narrowing the gap between Western and Chinese AI. The claim follows comments a day earlier from Treasury Secretary Scott Bessent, who told Fox Business the US could sanction foreign firms found to be stealing capabilities from American AI companies. The BBC has contacted Moonshot, Anthropic, the White House and Nvidia for comment, and the allegations come ahead of an expected September meeting between Trump and Chinese President Xi Jinping.
- White House adviser accuses Moonshot AI of stealing from Anthropic's model
- Moonshot allegedly used "distillation" and restricted Nvidia servers
- Comes as US weighs sanctions and before a planned Trump-Xi meeting