China AI Sanctions Loom as US Treasury Accuses Chinese Models of IP Theft

by Zaira Varghese 22. July 2026 AI 0

China AI sanctions moved from hypothetical to imminent this week after US Treasury Secretary Scott Bessent said Washington could penalize Chinese AI companies it believes built their models on stolen American intellectual property. Speaking on Tuesday, Bessent said the government is examining open-source models from China for signs of IP theft, as first reported by TechCrunch.

What Bessent Actually Said

“We are finding watermarks of our US large language models on many of the Chinese models, and that’s unacceptable,” Bessent told reporters. He described the technical practice at the center of the dispute — distillation — in blunt terms: “distillation, but you and I would call it theft.” Distillation is a legitimate machine-learning technique in which a smaller model learns from a larger one’s outputs, but US officials argue that some Chinese labs used it to clone proprietary American systems without permission.

Why China AI Sanctions Would Be a Big Deal

Model-level China AI sanctions would be unprecedented. To date, US restrictions on China’s AI ambitions have focused on hardware — export controls on Nvidia chips and manufacturing equipment. Targeting the models themselves would open an entirely new front, potentially barring US firms from using or distributing specific Chinese open-weight models and freezing the companies behind them out of the American financial system. That could ripple across the open-source community, where Chinese labs have become some of the most prolific releasers of powerful free models.

The Companies in the Crosshairs

The warning lands squarely on China’s leading open-model developers. Anthropic previously accused several Chinese firms of distilling its research, and the names that keep surfacing are familiar:

DeepSeek — the low-cost model maker whose valuation recently surged.
Moonshot AI — maker of the chart-topping Kimi K3.
MiniMax — another fast-rising Chinese lab.

All three have denied wrongdoing, and proving that a watermark constitutes theft rather than legitimate learning will be legally and technically contentious.

A Warning Shot, Not Yet a Policy

Importantly, no sanctions have actually been filed. Bessent’s comments amount to a public warning shot. Implementing model-level China AI sanctions would require the Treasury’s Office of Foreign Assets Control to write new designation criteria — a slow, complex process with no clear precedent. The Trump administration has said it supports open-source AI broadly, but will not tolerate foreign companies copying US technology, setting up a tension between openness and enforcement.

The Bottom Line

The threat of China AI sanctions marks a sharp escalation in the US–China technology rivalry, extending the battle from chips to the models those chips run. Whether Washington follows through or uses the threat as leverage, the message to Chinese labs is clear: the era of quietly training on American outputs may be ending.