Y Combinator’s Garry Tan wants U.S. open-weight AI labs to ‘distill’ frontier models, too

Details

Y Combinator CEO Garry Tan broke with frontier AI labs this week by arguing that distillation — where one model extensively prompts another to extract its reasoning patterns — should be permitted for U.S. open-weight AI developers. Asked about Chinese labs distilling American frontier models, Tan told CNBC he would 'do nothing' and called for a formal American distillation regime in an interview with TechCrunch.\n\nHis argument has two parts. First, he contends frontier labs lost the moral high ground when they trained on vast amounts of copyrighted human content without permission. Second, he argues that restricting how customers use API call outputs is an overreach — particularly when the underlying models were built on broadly public human knowledge.\n\nThis puts Tan in direct opposition to Anthropic CEO Dario Amodei, who has publicly called on U.S. regulators to crack down on distillation. Anthropic released its second report this week alleging Chinese labs are conducting 'illicit distillation attacks' using stolen credentials and fraud. Tan is not endorsing stolen credentials — he specifically wants distillation through legitimate API access to be normalised.\n\nThe deeper concern is market structure. Tan views a single proprietary lab monopolizing frontier AI as the real 'doomer scenario.' His preferred outcome combines frontier labs that drive capability advances with U.S. open-weight alternatives that keep AI access broad and prevent dangerous concentration of power.