Bessent Warns US May Sanction China Over AI IP Thef

The Trump administration just put Chinese AI developers on notice, and the trigger was a model that got too good too fast. Treasury Secretary Scott Bessent said Tuesday the United States could impose sanctions on China if evidence confirms Chinese AI companies trained their models using stolen American intellectual property, according to CNBC's reporting on the comments.

"If we see, especially that overseas models are stealing from our great companies, we have the ability to sanction them because of this theft," Bessent told Fox Business. He said the administration is specifically looking at whether Chinese AI models have been "distilled" from American models, a training method where a smaller, less capable model is built using outputs generated by an existing, stronger model.

The timing wasn't coincidental. Bessent's comments came just three days after Moonshot AI's Kimi K3 model attracted international attention for performance metrics that independent researchers say outpace what its publicly disclosed computing resources should theoretically produce, a discrepancy Eastern Herald's reporting noted as the specific spark behind Treasury's scrutiny. Kimi K3's rapid rise, including its recent scramble to pause new subscriptions after demand overwhelmed its capacity, has made it the highest-profile test case for these concerns yet, a story we covered in our piece on China's Kimi K3 halting new signups.

Bessent went further on the specifics. "We are finding watermarks of our U.S. large language models on many of the Chinese models, and that's unacceptable," he said, according to IBTimes's coverage of the interview, adding the administration would "be looking at that in the coming days or weeks." Anthropic itself sent a letter to the U.S. Senate Committee on Banking, Housing, and Urban Affairs last month alleging that Chinese tech company Alibaba carried out what it called "the largest known distillation attack" against it to date.

This warning arrives just as the U.S. and China prepare for their first formal AI talks under the current administration. Reuters reported Tuesday that the two countries plan to hold discussions in September, an outcome of the Trump-Xi summit held in May, with Bessent representing the U.S. side, according to CNBC's separate coverage of the planned dialogue.

The tension reflects a broader dynamic worth understanding. Chinese open-weight models have gained real traction against leading American offerings in recent months, a shift we've tracked closely in our DeepSeek statistics coverage since the original DeepSeek disruption in early 2025. Bessent's framing that "we are in the lead" has previously been his stated justification for why the U.S. can engage in these discussions from a position of strength rather than defensiveness.

I've advised companies on AI adoption for four years, and this story is worth watching closely if your business relies on any open-weight Chinese AI models, even indirectly through a vendor's infrastructure. Sanctions targeting specific Chinese AI companies could disrupt access to models your business or vendors currently depend on with very little advance warning, similar to how prior chip export restrictions reshaped hardware availability.

For businesses evaluating AI vendors right now, this is a reminder to understand exactly which underlying models power the tools you use, including any that route through open-weight Chinese systems, since geopolitical risk is becoming a genuine vendor selection criterion, not just a pricing or capability one.

Treasury Secretary Scott Bessent said the U.S. could sanction China if Chinese AI companies are found to have trained their models on stolen American intellectual property, specifically citing concerns about "distillation" from U.S. models. The warning came days after Moonshot AI's Kimi K3 drew scrutiny for performance that outpaces its disclosed compute resources. The U.S. and China are separately planning their first formal AI talks in September.