Y Combinator's Garry Tan Wants US Labs to Distill Too

The commander of Silicon Valley's top accelerator breaks with Anthropic — calling for an "American distillation regime" instead of a crackdown.

Published: 2026-09-13 Category: Quick Take Sources: TechCrunch

A Rare Split in the AI Consensus

Anthropic spent this week releasing its second report alleging Chinese labs are engaged in "illicit distillation attacks" — hiding their identities, relying on fraud and stolen credentials — and CEO Dario Amodei has publicly called on US regulators to crack down. You'd expect the Silicon Valley establishment to line up behind that. Y Combinator CEO Garry Tan is explicitly not doing so.

Asked about Chinese labs extracting knowledge from frontier models, Tan told CNBC, "I would do nothing. We could argue that there should be an American distillation regime." He elaborated to TechCrunch that he wants smaller, American open-weight labs to use the same distillation techniques on American frontier labs — building a domestic open-weight ecosystem that isn't dependent on Chinese models.

What Distillation Actually Is

Distillation is when a model maker extensively prompts another model to learn how it works and reasons. It's commonly and legitimately used to train new models. The dispute is over the illicit version — labs that distill without permission by concealing their identity and using stolen credentials.

Tan is careful to say he's not advocating for stolen credentials. His argument is about the front door. Two points drive it:

First, he sees it as an overreach for AI labs to dictate what customers can do with the information their models share. Second, he notes the proprietary labs didn't ask permission when they vacuumed up human knowledge — famously ingesting copyrighted material without asking. As Tan framed it: "Controlling what users and customers" can do with model outputs is a strange position for companies that scraped the internet wholesale.

The Strategic Logic

Tan's position is less contrarian than it first appears. From Y Combinator's vantage point, open-weight models are a startup force-multiplier: young companies can build on them without paying frontier tolls or handing data to incumbents. If the US cedes open-weight leadership to China, then every US startup's foundation is foreign. An "American distillation regime" would weaponize the same technique in service of domestic competition.

Amodei sees distillation as an attack on frontier IP. Tan sees it as standard practice that American labs should embrace. They're not even disagreeing about facts — they're disagreeing about who the technique should serve.

The Deeper Question

This is a policy fork hiding inside a technical debate. Do frontier companies get to control what happens with the knowledge they generate, the way a patent holder does? Or is frontier knowledge so derivative of collective human intelligence that distillation is fair game — as Tan implies, given the labs' own training-data history?

There's no clean answer, but the fact that the head of the most influential startup accelerator is arguing the open side is notable. It suggests the coming regulatory fights over distillation won't be a unified front.

Sources: TechCrunch, "Y Combinator's Garry Tan wants US open-weight AI labs to 'distill' frontier models, too" (Sep 11, 2026).