Garry Tan Advocates for U.S. 'Distillation Regime' to Counter Chinese AI Labs

Y Combinator CEO Garry Tan is calling for an "American distillation regime," arguing, as TechCrunch first reported, that U.S. open-weight AI labs should be permitted to use distillation techniques to extract knowledge from frontier models. Speaking with CNBC and TechCrunch, Tan suggested that regulators should not intervene to stop this practice, as he believes it would provide the U.S. with more robust open-weight options to compete with Chinese labs.

Distillation occurs when a model maker uses extensive prompting to learn how another model reasons. While common in the industry, the practice has become a point of contention. Anthropic recently released a report alleging that Chinese labs are conducting "illicit distillation attacks" by using stolen credentials and fraud to distill models without permission. Anthropic CEO Dario Amodei has publicly urged U.S. regulators to crack down on these activities.

Tan disagrees with Amodei's stance, asserting that it is an overreach for AI labs to control how customers use information provided via API calls. He further argued that proprietary labs did not seek permission when they ingested copyrighted material and human knowledge to train their own models. Consequently, Tan told TechCrunch that intelligence trained on broad public access data should be viewed as a "public good" rather than something restricted by terms of service.

While Tan supports the business model of frontier labs, he warned that the "doomer scenario" would be a monolithic market where a single company controls the best researchers and capital. He maintains that open-weight models are necessary to ensure freedom and access.