Y Combinator's Garry Tan says U.S. open-weight labs should distill frontier models too

Y Combinator CEO Garry Tan said in a recent CNBC interview that he believes regulators should stay out of the practice of distillation, even when Chinese AI labs use it to extract knowledge from leading American model makers. More than that, he argued U.S. labs should consider doing the same.

Distillation involves one model maker extensively prompting another company's model to learn how it works and reasons. It's a technique widely and legitimately used across the industry to help train new models.

Speaking to TechCrunch, Tan explained he wants smaller, American open-weight AI labs to apply the same training techniques to U.S. frontier labs, giving the country a broader set of open-weight options that aren't Chinese. He was careful to note he isn't endorsing stolen credentials or fraud — he wants companies to distill through legitimate means, essentially "through the front door."

A two-part argument

Tan's position rests on two points. First, he believes it's overreach for closed-model labs to dictate what customers can do with information the models provide them. Second, he notes that the same proprietary labs didn't ask permission when they scraped vast amounts of human knowledge — including copyrighted material — to train their own systems in the first place.

His comments come as Anthropic this week released its second report alleging that Chinese labs are conducting illicit distillation attacks, concealing their identities and relying on fraud and stolen credentials. Anthropic CEO Dario Amodei has previously called publicly for regulators to crack down on distillation, a position Tan doesn't share.

Tan said he wants balance between open-weight labs and frontier labs, arguing frontier labs need to remain a viable business while open-weight models give people freedom and access. He described the true worst-case scenario as all of AI's power concentrating in a single proprietary company.