Garry Tan backs open distillation by U.S. AI labs

Y Combinator CEO Garry Tan says U.S. open-weight AI labs should be free to distill frontier models, arguing that regulators should not block the practice.

Distillation involves extensively prompting another model to learn how it works and reasons. AI labs commonly and legitimately use the technique to help train new models. Tan told CNBC that he would do nothing about Chinese AI labs using distillation and said there could be an American distillation regime.

He later explained to TechCrunch that smaller American open-weight labs should be allowed to use similar training techniques on American frontier AI labs. In his view, that could create more robust open-weight options that are not Chinese.

Tan is not advocating stolen credentials or hidden identities. He wants American labs to be able to distill with permission. He argues that AI companies should not dictate what customers do with information shared through their models, and says access to intelligence trained on broadly available public data should be treated more like a public good.

Anthropic CEO Dario Amodei has called for U.S. regulators to crack down on unauthorized distillation, after Anthropic alleged that Chinese labs carried out illicit distillation attacks. Tan instead wants a balance: frontier labs should remain fundable businesses, while open-weight models provide freedom and access. He says the worst outcome would be one powerful proprietary company controlling the best AI.