Y Combinator CEO Garry Tan expressed his support for American open-weight AI labs utilizing knowledge distillation—a technique where a smaller model learns from a larger, more complex model—to build a robust domestic AI ecosystem. In an interview with CNBC and subsequent comments to TechCrunch, Tan suggested that U.S. regulators should not restrict these training methods, which he views as essential for maintaining American competitive advantages and preventing a monopoly by a single proprietary provider.

Tan's comments come as Anthropic released its September 2026 threat intelligence report, which details various malicious uses of its Claude models. The report highlights a trend of "illicit distillation attacks," where threat actors allegedly use stolen credentials or fraud to extract knowledge from frontier models without permission. Anthropic CEO Dario Amodei has previously called for increased regulation to crack against such practices.

While Tan is not advocating for the use of stolen credentials, he argues that restricting the use of API calls to study closed-weight models is overly constraining. He emphasized that access to intelligence trained on public data should be treated more as a public good than a proprietary secret. Tan noted that the "doomer scenario" for AI involves the concentration of immense power within a single, monolithic corporation.

Sources: