Y Combinator CEO Garry Tan Says ‘Do Nothing’ on AI Distillation Concerns

Y Combinator CEO Garry Tan has urged regulators to take no action regarding the practice of AI model distillation, dismissing calls for stricter controls despite heightened concerns from major technology firms and U.S. security agencies.

Speaking at Y Combinator’s annual Demo Day, Tan stated he would “do nothing” about the issue of OpenAI and Anthropic models being distilled by open-source competitors in China. While some Silicon Valley leaders and national security experts are demanding intervention, Tan argued that the focus should instead remain on tangible, near-term risks of artificial intelligence rather than catastrophic extinction scenarios dominating current headlines.

Distillation involves using the outputs of a more advanced AI model to train a smaller or less capable version, a technique that frontier developers say can undermine their intellectual property. Anthropic has accused Chinese companies such as Moonshot AI, DeepSeek, and MiniMax of employing the practice, while OpenAI has alleged that DeepSeek’s V3 and R1 architectures were derived from its GPT-4 and GPT-4o models.

The warnings come as the National Security Agency, the Cybersecurity and Infrastructure Security Agency, and the FBI released a joint cybersecurity advisory on Tuesday warning about China-based AI companies distilling U.S. frontier models.

Tan noted that critics of these complaints often point out that much of the data used to train AI models may fall under copyright protections. He highlighted ongoing legal battles, including The New York Times’ 2023 lawsuit against OpenAI and Microsoft over unauthorized use of articles, and a 2025 settlement between a consortium of book authors and Anthropic for $1.5 billion.

Instead of restricting distillation, Tan suggested that regulators should aim for a balance between open-weight models and frontier models, provided that frontier models maintain a price premium sufficient to sustain their business models. “This is actually the ideal case,” Tan said. “You want open weight models to give people freedom and access.”

Tan acknowledged that striking this balance is difficult, describing it as “a tightrope,” but believes it is a worthwhile pursuit that could lead to the best possible outcome for the industry. He also advocated for a more restrained approach to AI safety discussions.