In a startling development that could worsen U.S.-China tensions, Moonshot AI stands accused of covertly distilling Anthropic’s Fable model to develop its Kimi K3 system. This claim was made by White House Director Michael Kratsios, adding another layer of intrigue to the already competitive AI landscape.
The core of the accusation centers on Moonshot AI’s alleged use of sophisticated techniques to avoid detection while distilling Fable 5, one of the leading AI models developed by Anthropic. Such methods reportedly included routing access to advanced GPUs through third countries and leveraging internal platforms to extract large-scale capabilities.
The details
Kimi K3, which hit the market around July 16, 2026, boasts a staggering 2.8 trillion parameters and a context length of 1 million tokens. This makes it one of the most powerful AI models, directly competing with the likes of Fable 5 and OpenAI’s GPT 5.6 Sol. Notably, Kimi K3 has excelled in coding and agentic benchmarks, indicating substantial advancements over previous models.
Anthropic had already raised concerns in February 2026, when they accused Moonshot of engaging in 3.4 million Claude exchanges to distill Fable. This set the stage for the recent allegations, which have ramped up scrutiny surrounding Moonshot’s activities.
