Y Combinator CEO Garry Tan says the US government should stay out of the debate over AI model distillation. He shared his views in an interview with CNBC this week. Tan said regulators should let the practice continue rather than restrict it.
Distillation is a training method used across the AI industry. It happens when one company prompts another company's model many times. The goal is to learn how that model reasons and produces answers.
AI labs commonly use this technique to help train new systems. It is considered a normal part of building AI products. Tan believes the government has no reason to interfere with it.
What Tan Told CNBC
Tan told CNBC he would take no action against distillation. He said there could even be an argument for an American distillation approach. He explained to TechCrunch that he wants smaller US open-weight AI labs to use the same methods on bigger American frontier labs.
His goal is to give the US more open-weight AI options that are not built in China. Tan is not suggesting labs use stolen logins or fraud to do this. He wants companies to access model outputs through normal, legitimate use.
Tan's comments follow a new report from Anthropic. The report was released this week and covers what Anthropic calls illicit distillation attacks. Anthropic says Chinese labs have hidden their identities and used stolen credentials to distill without permission.
Anthropic CEO Dario Amodei had already asked US regulators to take action on distillation. He made those comments in an earlier public statement. Tan's position stands apart from that call for stricter oversight.
Tan's Argument For Open Access
Tan believes it is unfair for AI labs to control what customers do with information their models produce. He points out that big AI labs did not ask permission when they trained their own systems. Those systems were built using large amounts of public data, including copyrighted material.
He said access to intelligence trained on public data should work more like a public good. Tan argued this openness matters more than keeping information locked behind restrictive terms of service. He raised this point when TechCrunch asked why American labs should be allowed to distill freely.
Tan is known for heavy use of AI tools himself. He has described his own AI usage in dramatic terms in past interviews. He continues to push for a balance between open-weight labs and closed frontier labs.
He told CNBC that frontier labs deserve to remain profitable businesses. At the same time, he said open-weight models give people more freedom and access to AI tools. He sees both models playing a role in the industry.
Tan's biggest concern is a future where one company controls all frontier AI. He called this the doomer scenario for the AI industry. He said a single company with the best funding and researchers could end up running away with all the power.
Tan said that outcome would be bad for the wider AI field. His comments reflect an ongoing debate in the industry over how open AI development should be. The conversation comes as US and Chinese AI labs continue to compete on model performance and access.
