Moonshot AI K3 Model Faces Distillation Claims

Moonshot AI's K3 model is under scrutiny, with allegations surfacing that the Chinese firm distilled Anthropic’s Fable for its development. Michael Kratsios, a former White House CTO, stated that Moonshot AI employed a sophisticated internal platform to conduct large-scale distillation against U.S. models, designed to evade detection through rapidly changing access methods.

This platform reportedly allowed Moonshot AI to quickly switch between various access points, making it difficult to trace their activities. The company has also acquired GB300-equipped servers and accessed GB300s in Thailand, likely for training its AI models.

The U.S. government supports open AI development, including competitive ecosystems and legitimate distillation for creating efficient models. However, Kratsios emphasized that large-scale, covert industrial distillation, particularly when aimed at AI intellectual property theft and undermining American research, is unacceptable.

The specific model allegedly targeted for distillation is Anthropic Fable. These claims highlight escalating tensions over intellectual property in the global AI race.