started · updated
Moonshot AI accused of stealing Anthropic technology for Kimi K3 model
White House Science and Technology Policy director Michael Kratsios alleged that Beijing‑based Moonshot AI carried out a large‑scale “distillation” of Anthropic’s Fable model to develop its new Kimi K3 system. The claim includes that Moonshot built an internal platform to extract knowledge from U.S. models and that it accessed Nvidia GB300 servers, some reportedly located in Thailand, to train the model. Treasury Secretary Scott Bessent warned that such covert industrial‑scale distillation could trigger sanctions or placement on the U.S. Entity List. Moonshot has not publicly responded to the accusations, though it temporarily halted new subscriptions to Kimi K3 after demand surged. The Kimi K3 model has been reported to match or exceed performance of leading U.S. models such as Anthropic’s Claude and OpenAI’s GPT‑5 on certain benchmarks, prompting debate in the U.S. about AI policy and export controls. Nvidia CEO Jensen Huang described Chinese models as “excellent” and suggested that high‑quality open‑source AI could drive hardware sales, while also noting concerns about potential intellectual‑property violations.