Garry Tan: let U.S. open-weight labs distill frontier AI

Tan's position is unusual because it comes from the head of a startup accelerator that funds AI companies, not from a lab defending its own models. Anthropic's report this week described Chinese labs hiding their identities to distill without permission, relying on fraud and stolen credentials — and its CEO had already publicly asked regulators to crack down. Tan's answer is the opposite: leave it alone, and let American open-weight labs use the same techniques on American frontier models.

His argument rests on two claims. First, he says it's an overreach for AI labs to dictate what their customers can do with what their models share. Second, he notes the proprietary labs didn't ask permission when they ingested copyrighted material to train their own models, so controlling API calls on closed-weight models feels constraining.

Where this lands may hinge on whether regulators treat distillation as a boundary to police or a practice to normalize. Tan frames the stakes as a choice between fundable frontier labs and a single monolithic provider holding the best capital and researchers; he wants both to coexist. Which side U.S. policy takes is still undecided.