Y Combinator CEO Garry Tan is pushing back against efforts to restrict AI model distillation, arguing that U.S. regulators should allow American open-weight labs to use the same technique on frontier models.
Distillation is a process where one AI model is extensively prompted to extract knowledge about how another model works and reasons. According to Tan, American open-weight AI labs should be permitted to distill frontier American models through legitimate means, creating a more robust ecosystem of open-weight AI options that don’t rely on Chinese alternatives.
“I would do nothing,” Tan told CNBC. “We could argue that there should be an American distillation regime.” He elaborated that he wants smaller American open-weight labs to use training techniques on American frontier AI labs, emphasizing that he is not advocating for theft or fraud—only that labs should be free to come “in the front door.”
Tan’s position contrasts sharply with that of Anthropic. The AI safety company released a report this week alleging that Chinese labs are engaged in “illicit distillation attacks,” using fraud and stolen credentials to distill without permission. Anthropic CEO Dario Amodei has previously called on U.S. regulators to crack down on the practice.
Tan’s argument rests on two pillars. First, he contends that proprietary AI labs overreach when they dictate what customers can do with information their models share via API calls. Second, he notes a historical parallel: frontier labs did not ask permission when they trained their models on vast amounts of human knowledge, including copyrighted material.
“Controlling what users and customers do with API calls to closed weight models feels constraining,” Tan told TechCrunch. He suggested government has a role in “normalizing 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’s broader vision is to maintain balance between open-weight and frontier AI labs. He acknowledges that frontier labs “are at the frontier and driving it forward” and deserve sustainable business models. However, he emphasizes the importance of open-weight models providing “freedom and access.”
His ultimate concern is concentration of power. “The nightmare scenario, the doomer scenario for AI is that there’s just one company,” he said, arguing that a monopoly on frontier AI would be damaging. Tan sees supporting American open-weight distillation as a way to prevent such consolidation and preserve competitive dynamism in the AI market.
Key facts
- Garry Tan believes U.S. open-weight AI labs should be allowed to distill frontier American models through legitimate means
- Distillation involves extensively prompting a model to learn how it works and reasons
- Anthropic released a report alleging Chinese labs conduct illicit distillation using fraud and stolen credentials
- Tan argues that proprietary AI labs should not control what customers do with API information
- Frontier labs trained their models on vast amounts of data, including copyrighted material, without seeking permission