Garry Tan Opposes Restrictions on AI Distillation, Calls for Establishing an "American Distillation Framework"

Y Combinator CEO Garry Tan recently stated that US regulators should not impose comprehensive intervention on AI model distillation and could even consider establishing an "American distillation ecosystem." He hopes smaller American open-weight labs can leverage domestic frontier models to train their own systems under legitimate API access, thereby expanding open choices beyond those offered by Chinese models. This week, Anthropic released a report stating it detected and blocked unauthorized distillation activities originating from seven Chinese laboratories. The operators were accused of bypassing restrictions through fake identities, stolen credentials, and proxy accounts. Earlier this year, Anthropic CEO Dario Amodei also called on regulators to crack down on industrial-scale distillation. Tan emphasized that he does not support account theft or credential misuse but opposes over-restrictive measures by model providers regarding how paying customers utilize API outputs. Tan pointed out that many closed-source models absorb substantial amounts of public and copyrighted human works during training without obtaining individual authorizations; therefore, locking model outputs entirely within service terms is unreasonable. He believes frontier labs must maintain sustainable business models, while open-weight models should provide developers with freedom and choice. More than concerns about distillation itself, he worries that capital, talent, and advanced AI capabilities may ultimately concentrate in a single closed entity.