US threatens sanctions against Chinese AI models over IP theft
TechCrunch Rebecca Bellan ● Covered by 75 sources
Treasury Secretary Bessent says the US may sanction Chinese AI firms if their open models are found stealing US company IP. Chinese models like Kimi K3 are catching up fast, and Washington's worried about losing the AI race.
Based on reporting by TechCrunch, Rebecca Bellan — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Scott Bessent went on Fox Business Tuesday and said something that should make Beijing's AI labs nervous: the Treasury is going to start scrutinizing open source models coming out of China for signs of stolen intellectual property, and sanctions are on the table if they find it. He was careful to frame it as pro-open-source, anti-theft, not a blanket attack on Chinese AI. But the timing tells its own story.
Moonshot AI just released Kimi K3, and it's good enough that people in Silicon Valley are once again asking uncomfortable questions about whether OpenAI and Anthropic can keep raising money at their current valuations if a free Chinese alternative keeps closing the capability gap. Axios reported Monday that the administration is even weighing an outright ban on Chinese open models, though that report has been pushed back on by other sources. Either way, this is clearly the next front after chip export controls: if you can't stop China from building good models, maybe you go after the models themselves.
The technical crux here is distillation, the process of squeezing a big model's knowledge into a smaller, cheaper one. Whether that counts as theft is genuinely contested. Satya Nadella pointed out the irony of large labs training on the entire public internet and then crying foul when someone distills their outputs. Hugging Face's Clem Delangue went further, arguing distillation is a minor factor at best and that China's progress comes down to strong research teams working more openly and collaboratively than their American counterparts — not IP theft.
Meanwhile the US labs pushing hardest for this crackdown aren't exactly spotless themselves. Anthropic just got court approval to start paying out its $1.5 billion settlement after a judge found it illegally downloaded millions of copyrighted books to train its models. That context makes Bessent's IP-theft framing land a little awkwardly, even if the underlying competitive anxiety driving it is entirely real.
My take — AI-written commentary, not fact-checked reporting
Calling distillation theft while your own industry just settled for illegally hoovering up copyrighted books is rich, and I think this is less about IP and more about protecting valuations that assumed China would stay years behind. Sanctioning open models sets a precedent that could just as easily boomerang back onto US labs' own training practices, and honestly, if Kimi K3 is winning on merit, tariffs on math won't fix that.
Read more about this at: TechCrunch