The United States government is prepared to impose sanctions on Chinese artificial intelligence models if evidence confirms intellectual property theft from American companies. Treasury Secretary Scott Bessent announced on Tuesday that the administration will scrutinize open-source AI models originating from China for violations, stating that the U.S. has the ability to sanction them if theft is proven. This warning comes amid growing competition in the artificial intelligence sector and follows scrutiny of Moonshot AI's Kimi K3 model for performance metrics that some researchers believe exceed its disclosed computational resources.
Bessent articulated the Treasury's specific concern: that Chinese AI firms may be incorporating proprietary elements, such as model weights or training outputs, from American frontier models into their own products. This practice, often referred to as "distillation," allows competing models to achieve advanced capabilities without the extensive research and development investment made by U.S. companies. The Treasury Secretary noted that the U.S. government has detected watermarks of American large language models on numerous Chinese AI models, which he deemed unacceptable.
The potential sanctions represent an escalation of the Trump administration's efforts to slow China's advancement in artificial intelligence, building on previous measures such as restrictions on advanced semiconductor chips and export controls. While the administration supports open-source AI development, Bessent emphasized a firm stance against intellectual property theft. The scale of alleged IP theft is considerable, with an April 2026 Senate Judiciary Committee hearing estimating annual U.S. losses between $400 billion and $600 billion due to Chinese IP theft.
The timing of Bessent's announcement is notable, occurring shortly before scheduled government-level AI discussions between the U.S. and China in September. This move functions as both a negotiating tactic and a policy signal, indicating a potential shift from targeting hardware to directly addressing AI software and training practices.
The debate over distillation is complex, with some in the industry questioning whether it constitutes theft or a common development method. Microsoft CEO Satya Nadella has suggested that major AI companies utilize fair use for training on public data while restricting others from distilling their systems. Similarly, Hugging Face CEO Clem Delangue has attributed China's AI progress to strong research teams and an open development approach, with distillation playing a minor role.
This potential regulatory action could have implications for the global tech and crypto ecosystems, particularly for decentralized AI platforms that integrate or fine-tune Chinese models. Companies and individuals may face compliance challenges if Chinese models are sanctioned, potentially requiring migration to U.S.-approved alternatives and leading to increased development costs and delays. The upcoming U.S.-China AI talks will be a key event to monitor for further developments in this escalating technological competition.