Garry Tan, CEO of Y Combinator, is pushing back against efforts to restrict how AI labs use distillation techniques to learn from competitor models. In a recent interview, he told CNBC he believes regulators should not intervene when Chinese labs use distillation, and that the U.S. should develop its own “distillation regime” instead.
Distillation is a legitimate AI training technique where one model maker extensively prompts another model to understand how it works and reasons. Tan wants smaller American open-weight AI labs to apply distillation to frontier models created by U.S.-based companies, generating a more robust ecosystem of open-weight options that compete with Chinese alternatives.
His stance contrasts sharply with recent warnings from Anthropic. The AI safety company released a report this week alleging that Chinese labs are conducting “illicit distillation attacks,” using fraudulent identities and stolen credentials to extract knowledge without permission. Anthropic CEO Dario Amodei has publicly called on U.S. regulators to crack down on such practices.
Tan clarified that he is not advocating for American labs to use stolen credentials or fraudulent methods. Instead, he argues that AI labs should allow legitimate users to access their models openly. His position rests on two key arguments.
First, he contends that proprietary AI labs are overreaching by trying to control what customers do with information their models share through API calls. Second, he points out that frontier labs themselves did not ask permission when training on vast amounts of human knowledge, including copyrighted material, without compensating or seeking consent from intellectual property holders.
“Controlling what users and customers do with API calls to closed weight models feels constraining,” Tan told TechCrunch. He argued that access to intelligence trained on broadly available public data should function as a public good rather than being locked behind restrictive terms of service.
Tan’s broader vision emphasizes balance between open-weight and frontier AI labs. He supports continued funding and business viability for frontier labs driving innovation forward, while also advocating for open-weight models to provide freedom and broader access.
According to Tan, the worst-case scenario for AI development would be consolidation of immense power in a single company. “The nightmare scenario, the doomer scenario for AI is that there’s just one company,” he 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.”
Key facts
- Tan told CNBC he would “do nothing” to restrict distillation and suggested the U.S. develop an “American distillation regime”
- Distillation is a legitimate technique where models learn from other models through extensive prompting
- Anthropic this week released a report alleging Chinese labs conduct “illicit distillation attacks” using fraud and stolen credentials
- Tan argues proprietary labs should not control what customers do with API data, noting labs themselves trained on copyrighted material without permission
- Tan believes consolidation of AI power in a single company represents the worst-case scenario