Y Combinator CEO urges U.S. open-weight AI labs to freely distill frontier models

Y Combinator CEO Garry Tan has publicly argued that U.S. regulators should refrain from restricting American open-weight AI labs from distilling frontier models, positioning the practice as a strategic counterbalance to proprietary dominance in AI. In an interview with CNBC, Tan stated, “I would do nothing,” when asked about regulating distillation, adding that the U.S. could establish an “American distillation regime” to normalize the practice. He elaborated to TechCrunch that this would allow smaller, open-weight labs to extract knowledge from frontier models, thereby expanding the availability of U.S.-developed open-weight options and reducing reliance on Chinese alternatives. Distillation is a technique where a model is extensively prompted to learn the reasoning and capabilities of another model, and it is commonly used by AI labs to train new systems.

Tan’s position contrasts with recent allegations from Anthropic, which accused Chinese labs of conducting “illicit distillation attacks” by using stolen credentials and fraud to access frontier models without permission. Anthropic CEO Dario Amodei has publicly urged U.S. regulators to crack down on such practices. Tan, however, does not advocate for illicit access; instead, he argues that restricting distillation would be an overreach, particularly given that frontier labs have historically ingested vast amounts of public data—including copyrighted material—without explicit permission to train their models.

He framed access to intelligence trained on broadly accessible public data as a public good, stating that controlling what users can do with API calls to closed-weight models feels “constraining.” Tan also emphasized the need to maintain a balance between open-weight and frontier labs, asserting that frontier labs drive innovation forward while open-weight models provide freedom and access. “They are at the frontier and driving it forward,” he told CNBC. “We want that to be fundable, and be a great business model ongoing. You want open weight models to give people freedom and access.”

Tan warned of a “doomer scenario” where a single proprietary provider could become “monolithic,” consolidating power and stifling competition. “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.” His remarks underscore a broader debate about the future of AI development, competition, and the role of regulation in shaping an open versus closed ecosystem.