Bessent warns Chinese AI firms of sanctions over IP theft
Treasury Secretary Scott Bessent warned that Chinese AI firms engaging in industrial-scale distillation attacks amounting to IP theft face potential sanctions and Entity List designations. The White House alleged Moonshot AI used Anthropic’s Fable models to develop its Kimi K3 model and accessed Nvidia GB300 chips. Nvidia CEO Jensen Huang defended distillation as fundamental to intelligence, while Anthropic previously accused Alibaba's Qwen unit of a similar large-scale extraction campaign.

*this image is generated using AI for illustrative purposes only.
Treasury Secretary Scott Bessent warned that Chinese artificial intelligence companies could face sanctions and Entity List designations if they use "industrial-scale distillation attacks" that cross into intellectual property theft. Bessent stated on Wednesday that President Donald Trump’s administration supports open-source AI and the innovation it unlocks but clarified that "open source is not open season on American IP." He emphasized that when firms from the People's Republic of China conduct covert, industrial-scale distillation attacks that cross the line into IP theft, sanctions and Entity List designations will be on the table.
US Allegations Against Moonshot AI
White House science chief Michael Kratsios alleged that Moonshot AI used Anthropic’s Fable models to develop its Kimi K3 model through large-scale distillation. Kratsios claimed Moonshot AI built a sophisticated internal platform to conduct large-scale distillation against U.S. models, allowing them to quickly switch between multiple methods of access to avoid detection. He further alleged that Moonshot AI acquired or accessed NVIDIA Corp GB300-equipped servers in Thailand to train its AI models. Anthropic reportedly traced 3.4 million exchanges with its models to Moonshot AI, routed through hundreds of fake accounts.
Industry Reaction and Previous Accusations
NVIDIA Corp CEO Jensen Huang offered a contrasting perspective, stating that American companies should "absolutely" use Chinese AI models. "Distillation, learning from AI, learning from other sources of knowledge, is fundamental to intelligence," Huang told Axios. He argued there is "no scenario where China runs U.S. companies off the road" and suggested that cheap open models expand AI's audience, driving demand for chips and data centers.
Separately, Anthropic in June accused Alibaba Group Holding Ltd’s Qwen AI unit of conducting a large-scale "distillation" campaign to extract capabilities from its Claude models. The company alleged the effort relied on nearly 25,000 fraudulent accounts that generated more than 28.8 million interactions with Claude to help improve Chinese AI systems.
| Entity | Allegation | Key Details |
|---|---|---|
| Moonshot AI | Distilled Anthropic's Fable | Used internal platform to switch access methods; accessed Nvidia GB300 chips in Thailand |
| Alibaba Group Holding Ltd (Qwen) | Distilled Anthropic's Claude | Used ~25,000 fraudulent accounts; 28.8 million interactions traced |
How will the US distinguish between legitimate AI learning and illegal industrial-scale distillation when enforcing new sanctions?
What impact will potential Entity List designations have on the global supply chain for NVIDIA chips and data center infrastructure?
Could the threat of sanctions drive Chinese AI firms to accelerate the development of fully domestic hardware and software ecosystems?

























