The U.S. Treasury Department has threatened sanctions against Chinese artificial intelligence company Moonshot AI. This action follows White House accusations that Moonshot AI distilled Anthropic's Fable model to develop its Kimi K3 system. Treasury Secretary Scott Bessent stated that sanctions and Commerce Department Entity List designations are possible for Chinese firms engaging in "covert, industrial-scale distillation attacks" against American AI models.
Michael Kratsios, director of the White House Office of Science and Technology Policy, publicly stated that the administration possesses information indicating Moonshot AI developed "a sophisticated internal platform to conduct large-scale distillation against U.S. models". This platform allegedly allowed Moonshot AI to rapidly switch between access methods to evade detection. Kratsios also raised concerns that Moonshot AI acquired servers equipped with Nvidia GB300 chips and accessed similar systems in Thailand, potentially violating U.S. export restrictions on advanced AI hardware. Nvidia's GB300 servers are part of the company's Blackwell generation of AI training infrastructure, which is subject to U.S. export controls aimed at preventing China from acquiring advanced AI technology.
Knowledge distillation is a machine learning technique where a smaller AI model learns from the outputs of a larger, more complex model. While this process is common in the AI industry for creating more efficient models, the White House's objection centers on the alleged unauthorized use of proprietary outputs from Anthropic's Fable model for training purposes. Anthropic had previously reported that Moonshot AI generated millions of fraudulent conversations targeting its Claude model, which it claims were used for unauthorized training.
Anthropic released its Claude Fable 5 model, a publicly available version of its Mythos-class AI, in June 2026. This model is designed for complex reasoning and agentic tasks. Moonshot AI unveiled its Kimi K3 model shortly after Fable 5's re-release, leading to questions about the rapid development timeline. Some AI experts have questioned the technical feasibility of achieving Kimi K3's performance level within such a short period, even with distillation.
Treasury Secretary Bessent emphasized that while the administration supports open-source AI development, it will not tolerate intellectual property theft. He indicated that U.S. authorities are finding "watermarks" on Chinese models, suggesting the use of U.S. large language model outputs. Bessent stated that if foreign models are found to be stealing from American companies, sanctions are a viable recourse.
The accusations have intensified a broader debate in Washington regarding the influx of Chinese open-weight AI models and the U.S. government's response to perceived intellectual property theft and violations of export controls. Moonshot AI has not issued a public response directly addressing these allegations. The Chinese embassy in Washington has also not commented on the matter.