White House accuses Moonshot AI of illicit model distillation

Kratsios claimed on July 22 that the U.S. government holds information suggesting Moonshot AI distilled Anthropic’s Fable model to accelerate K3’s development. According to the official, the company engineered a sophisticated internal platform designed to extract knowledge from American models on an industrial scale. This system allegedly allowed Moonshot to rotate access methods rapidly, complicating detection efforts by U.S. developers.

While model distillation is a standard practice for creating efficient software, Kratsios distinguished between legitimate innovation and the covert appropriation of protected research. He characterized the alleged actions as an attempt to undermine American technological leadership. Moonshot AI has yet to issue a formal response to the accusations, and the White House has not released technical evidence to support the claims. The controversy arrives as the K3 model, promoted as a 2.8-trillion-parameter system, outperformed several U.S. counterparts in recent coding evaluations. Former White House official David Sacks warned that the situation highlights the risks of stifling domestic innovation through heavy regulation, which he argues may inadvertently allow Chinese competitors to close the performance gap.