The AI race just got a policy grenade
Treasury Secretary Scott Bessent says the U.S. is willing to go after Chinese AI models if they’re found to be built on stolen American IP. In other words: open-source is fine, but “copy-paste with extra steps” is not exactly getting the White House seal of approval.
He framed the issue around AI “distillation,” the practice of training a weaker model using outputs from a stronger one. Bessent said the administration has seen what he called watermarks from U.S. large language models showing up in Chinese systems, and that’s apparently the kind of thing that could trigger sanctions.
Why investors should care
This isn’t just diplomatic chest-thumping. If Washington starts tightening the screws, it could ripple across:
- Chinese AI startups trying to scale fast
- U.S. AI firms worried about model theft and export leakage
- Semis and cloud infrastructure names that sit in the middle of the U.S.-China tech chess match
Alibaba gets pulled into the crossfire
Alibaba shows up in the story because Anthropic previously accused the company and its Qwen AI unit of running a huge distillation campaign to extract capabilities from Claude. That doesn’t mean Alibaba is the whole story here — but it does mean the company is sitting uncomfortably close to the blast radius if sanctions talk turns into actual policy.
Big picture: AI is becoming the new trade war battleground, and the rules are getting made in real time. If you own companies exposed to U.S.-China tech flows, this is one of those headlines worth keeping on your radar.