White House Official Accuses Moonshot AI of Distilling Anthropic's Fable for Kimi K3; Nous Research Co-founder Retorts

According to White House Chief Technology Officer Michael Kratsios, Moonshot AI trained its Kimi K3 model using Anthropic's Fable, monitored by Beating on July 23. Kratsios alleged that Moonshot constructed an internal platform to batch-call U.S. models while evading detection, though no public evidence was provided. Moonshot has not yet responded.

Nous Research co-founder Teknium countered by replicating Kratsios's phrasing, claiming Anthropic scraped his GitHub, X, Reddit, Discord, and Facebook posts to train Fable. He added in sarcasm: "They are essentially distilling all people on Earth." Teknium highlighted a perceived double standard: U.S. models absorb global internet content as standard "training," while Chinese models learning from U.S. models face accusations of "theft."