Y Combinator CEO Garry Tan Urges U.S. Open-Weight Labs to Mirror Chinese Distillation Practices

Y Combinator CEO Garry Tan is calling on U.S. regulators to allow domestic open-weight artificial intelligence labs to utilize distillation techniques, mirroring methods currently employed by Chinese competitors. In an interview with CNBC, Tan stated he would take no action to restrict the practice and suggested that Washington should consider establishing a formal distillation framework for American companies.

Tan clarified to TechCrunch that his position involves smaller, U.S.-based open-weight labs applying similar training methods to their own frontier model providers. This approach, he argued, would expand the availability of open-weight options within the United States and reduce reliance on Chinese technology.

Distillation is a standard training technique wherein a developer prompts a model extensively to understand its reasoning processes. However, tensions have risen after Anthropic released a report this week alleging that Chinese labs are conducting “illicit distillation attacks.” The company claims these actors are concealing their identities and using stolen credentials or fraud to extract knowledge without permission. Anthropic CEO Dario Amodei had previously urged U.S. regulators to crack down on such activities.

Tan, the head of Silicon Valley’s most prominent startup accelerator, disagrees with calls for stricter oversight. He emphasized that he is not advocating for the use of unauthorized credentials, but rather for labs to access frontier models through legitimate channels. His argument rests on two main points: first, that major AI labs have no right to dictate how customers utilize the data their models generate, and second, that proprietary firms themselves ingested vast amounts of copyrighted material without permission during their own training phases.

“Controlling what users and customers do with API calls to closed weight models feels constraining,” Tan told TechCrunch. “There’s a role government can play here to normalize the fact that access to intelligence that was trained on broad public access data should itself also be more a form of a public good than something locked away behind restrictive terms of service.”

Tan, who recently described himself as having “cyber psychosis” due to his intensive use of AI tools, believes a healthy ecosystem requires a balance between open-weight and frontier laboratories. He acknowledged that frontier labs drive innovation and need viable business models, but stressed that open-weight models are essential for providing freedom and access to users.

He warned that the worst-case scenario for AI development is a monopoly held by a single proprietary provider. “The nightmare scenario, the doomer scenario for AI is that there’s just one company,” Tan said. “It has the best access to capital. It has the best AI researchers. It runs away with it and suddenly there’s one company that’s monolithic. And that would be bad.”