U.S. Treasury Secretary: Will Review IP Issues of Chinese AI Models
N.R. Finch
Treasury Secretary Scott Bessent says the US has found watermarks from American large language models inside multiple Chinese AI models and will launch a probe within days or weeks — signaling that Chinese open-source AI faces a systematic squeeze in the US market.
What exactly did Bessent say?
Bessent told Fox Business the administration supports open-source models but does not support IP theft.
He said the government has already found watermarks — hidden markers embedded in model output to prove origin — from US large language models inside "many Chinese models," calling it "unacceptable."
This means → Washington claims it already has preliminary evidence, not just suspicion. A formal investigation is set to begin within days or weeks.
What happens if the findings hold up?
Bessent explicitly raised sanctions: if overseas models are found stealing US companies' work, Treasury has the power to sanction them.
He floated a second track — mandatory disclosure: US companies using Chinese AI models may have to tell their customers.
In plain terms = one path cuts off market access for Chinese models directly; the other forces companies that use them to come clean. Both shrink the viable space.
What does this mean for the market?
Bessent compared Chinese AI models to "counterfeit goods." This reflects a regulatory framing that is shifting from tech competition toward IP enforcement.
If sanctions and mandatory disclosure advance in parallel, they will impose real constraints on the commercialization of Chinese AI models in the US.
This means → US firms relying on Chinese open-source models need to reassess supply-chain risk, while Chinese AI companies' path to the US market — already narrow — is being policy-blocked further.
Content is for reference only, not financial advice.