NewsSep 12, 2026, 11:07 UTC

Garry Tan calls for broader access to frontier-model knowledge

What’s the view? In a September 2026 interview with TechCrunch, Garry Tan, President and CEO of Y Combinator, argued that smaller U.S. open-weight AI labs should be able to use distillation techniques on frontier models. He said he would prefer regulators to “do nothing” about the practice, while distinguishing legitimate access from the illicit activity alleged by Anthropic in reports about Chinese AI labs using stolen credentials and concealed identities.

Why it matters: Tan’s argument is that capable AI models were trained on broad public human knowledge, and that access to the intelligence produced from that knowledge should be treated more like a public good than a resource locked behind restrictive terms of service. He wants American open-weight developers to have a lawful route to learn from frontier systems, helping create a stronger domestic ecosystem of models that are not controlled by a single proprietary provider.

Tan also stressed that the goal is not to legitimise fraud or credential theft. His position is instead that frontier AI companies should not be able to dictate how customers use information returned through their APIs, particularly when those companies themselves trained models on large bodies of publicly available and copyrighted material. He described a market dominated by one well-capitalised provider with the strongest researchers and best access to capital as the “doomer scenario”, and presented open-weight models as a counterbalance that could preserve freedom and access while leaving frontier labs with a viable commercial model.