Garry Tan Says Regulators Shouldn’t Curb Alleged Model Distillation

Garry Tan wants regulators to leave alleged copying of leading AI models alone. The Y Combinator CEO said he would “do nothing” about allegations that Chinese competitors distilled OpenAI and Anthropic models, arguing that policy should protect open-weight access while allowing frontier systems to retain a commercial premium.

The position puts Tan at odds with calls to curb the practice from frontier-model developers. Rather than treating the allegations as a reason for regulatory intervention, he said policymakers should pursue conditions that give people access to open-weight models while preserving the economic case for building the most advanced systems.

I would do nothing.

Garry Tan, Y Combinator CEO, speaking to CNBC

Access and incentives on a tightrope

Tan’s proposed equilibrium has two goals that can pull in different directions. He said open-weight models should provide freedom and access, but frontier-model developers need a price premium for their business model to remain feasible. He described striking that balance as a “tightrope.”

That prescription does not resolve the accusations that prompted it. Distillation uses the outputs of a more capable AI model to train a smaller or less capable one, and CNBC reported that the technique can sometimes be illicit. Tan’s position is about what regulators should prioritize, not a finding on whether a particular company committed misconduct.

The allegations behind the dispute

• Anthropic has accused Moonshot AI, DeepSeek and MiniMax of model distillation.
• OpenAI believes DeepSeek’s V3 and R1 architectures were distilled from GPT-4 and GPT-4o.

A different definition of urgent risk

Tan paired his distillation stance with a narrower view of AI safety policy. He said policymakers should respond to current evidence rather than science-fiction scenarios and identified cybersecurity as an immediate concern. His argument is not that AI risks can be ignored; it is that policy attention should center on threats he sees as already close at hand.

The policy choice Tan describes is therefore broader than a dispute between AI companies. It asks whether rules should focus on restricting alleged extraction of frontier capabilities, or on sustaining a market in which access spreads while the highest-end models can still command a premium. Tan argues for the latter approach.