The White House's chief science advisor, Michael Kratsios, has accused Chinese firm Moonshot of building its Kimi K3 model by illicitly copying Anthropic's Fable LLM and using chips barred from export to China, calling it "unacceptable" industrial distillation. The claim comes amid wider tension over whether Chinese open-weight AI models should be banned, but AI researchers contacted by TechCrunch say the timeline and technical evidence make simple distillation an implausible explanation for K3's strength, complicating the narrative behind the accusations.

Treasury Secretary Scott Bessent had separately claimed to have found "watermarks" of US models in Chinese systems, though neither he nor Kratsios detailed the evidence, and Moonshot did not respond to questions. Researchers Braden Hancock (Laude Institute/Snorkel AI) and Nathan Lambert (Allen Institute for AI) argued that copying Fable's capabilities so quickly and cheaply is implausible: Fable has only been public since 1 July, distillation via supervised fine-tuning is losing relevance as models near the frontier, and truly replicating advanced performance would require costly reinforcement learning involving tens of millions of agent interactions—something Anthropic's API would make prohibitively slow and expensive. Anthropic has separately accused Moonshot, DeepSeek and MiniMax of systematic distillation earlier this year, so some influence from frontier models remains plausible, even if it doesn't fully explain K3's capabilities.