The White House and U.S. officials have accused Chinese company Moonshot of illicitly replicating Anthropic’s Fable large language model (LLM) to create its own Kimi K3 model, which is currently the largest open-weight LLM available. White House science advisor Michael Kratsios claimed that Moonshot copied Fable’s capabilities using advanced Nvidia Grace Blackwell 300 GPUs, which are banned for export to China. Treasury Secretary Scott Bessent added that stolen U.S. AI intellectual property is a serious concern, with sanctions threatened for covert large-scale distillation attacks that infringe on proprietary technology.

Despite these allegations, AI experts expressed skepticism that Kimi K3’s rapid emergence and advanced performance could be explained solely by distillation of Fable, which was only publicly released on July 1. Researchers like Braden Hancock of the Laude Institute pointed out that distillation and supervised fine-tuning typically take much longer, making it implausible for Moonshot to have matched or surpassed Fable’s strength in a matter of weeks. Nathan Lambert from the Allen Institute for AI further noted that effective replication of cutting-edge LLMs now requires complex reinforcement learning methods, which entail vast computational resources and cannot be efficiently done via API access.

Anthropic previously accused Moonshot and other Chinese firms of systematic distillation attempts earlier this year, identifying millions of interactions consistent with capability extraction rather than normal usage. Although distillation is a common practice industry-wide—including by U.S. players like Elon Musk’s SpaceXAI—the scale and secrecy of Moonshot’s alleged efforts, combined with its access to black-market advanced GPUs forbidden for export, have elevated concerns over national security and intellectual property theft. The situation has intensified broader debates in Washington about restricting Chinese models and implementing stricter know-your-customer rules for data centers housing high-end AI hardware.

Nevertheless, experts emphasize that Chinese AI teams, including Moonshot, possess substantial technical expertise and are not merely relying on stolen models to progress. The controversy has spotlighted challenges in balancing open AI research with protecting innovation and security, especially as frontier AI capabilities rapidly evolve. Moonshot has yet to respond to inquiries about its training processes, while U.S. authorities continue to investigate potential violations related to export controls and unauthorized use of proprietary technology in AI development.