YC's Garry Tan backs US open-weight distillation of frontier AI
Original: Y Combinator’s Garry Tan wants US open-weight AI labs to ‘distill’ frontier models, too
Why This Matters
YC's stance could shape policy debates on distillation rights and open-weight AI competitiveness globally.
Y Combinator CEO Garry Tan said this week he wants smaller American open-weight AI labs to freely distill knowledge from frontier models, pushing back against calls by Anthropic's Dario Amodei for regulators to restrict the practice. Tan argues frontier labs have no standing to police API outputs, given their own unlicensed training data practices.
In a CNBC interview and follow-up comments to TechCrunch, Garry Tan laid out a position that puts him at odds with Anthropic CEO Dario Amodei: regulators should not crack down on AI distillation, and American open-weight labs should actively use the technique on domestic frontier models. 'I would do nothing,' Tan said of proposed restrictions. 'We could argue that there should be an American distillation regime.'
Distillation involves extensively prompting a model to learn how it reasons, and is widely used across the industry. Anthropic this week published its second report alleging Chinese labs are conducting 'illicit distillation attacks'—using stolen credentials and fraudulent identities to distill without permission. Amodei has publicly urged U.S. regulators to act.
Tan is not advocating for stolen credentials. His argument is that frontier labs should not dictate what customers do with API outputs—especially since those same labs trained on vast swaths of copyrighted human knowledge without asking permission. 'Controlling what users and customers do with API calls to closed-weight models feels constraining,' he told TechCrunch, adding that intelligence trained on broadly public data should be treated more as a public good than a locked proprietary asset.
Tan's broader goal is balance: frontier labs should remain fundable and commercially viable, while open-weight models give users freedom and access. His real concern is concentration—all frontier AI power consolidating in the hands of a single entity.