Y Combinator CEO Garry Tan Backs U.S. AI Distillation of Frontier Models
Y Combinator CEO Garry Tan backs legal AI distillation by U.S. open-weight labs, opposing tighter rules despite Anthropic's latest claims against Chinese labs.
Summary
In comments published September 11, 2026, Y Combinator CEO Garry Tan said regulators should leave Chinese AI distillation alone and could establish an American distillation regime. He wants smaller U.S. open-weight labs to extensively prompt American frontier models, learning how they operate and reason through a common, legitimate training method. Anthropic's second report that week alleged Chinese labs concealed their identities and used fraud and stolen credentials to conduct illicit distillation without permission. Anthropic CEO Dario Amodei has urged U.S. regulators to crack down.
Tan rejects stolen access but says frontier labs should not dictate how customers use knowledge returned through API calls. He argues proprietary developers trained their models on broad public knowledge, including copyrighted material used without rights holders' permission, and says government should treat resulting intelligence more like a public good than something restricted by terms of service. Tan wants frontier development to remain fundable while open-weight models preserve access and freedom. His worst outcome is one proprietary company combining the most capital, researchers and capable AI into a monolithic provider.
Positives
- Tan's proposed American distillation regime could give users more U.S. open-weight alternatives to Chinese models.
- Distillation is already a common, legitimate method for training models when access is authorized.
- Tan explicitly excludes fraud and stolen credentials from the distillation practices he supports.
- Frontier labs would remain fundable under Tan's proposed balance between proprietary innovation and open-weight access.
Risks & concerns
- Anthropic's second report alleges Chinese labs concealed their identities to distill models without permission.
- Fraud and stolen credentials were allegedly used in the illicit distillation attacks described by Anthropic.
- Anthropic CEO Dario Amodei wants a regulatory crackdown, directly conflicting with Tan's preferred hands-off approach.
- Tan warns that one capital-rich proprietary provider could monopolize leading researchers and frontier AI capabilities.