Y Combinator CEO Garry Tan Rejects AI Distillation Ban

Garry Tan, the chief executive of startup accelerator Y Combinator, has argued that regulators should take no action against AI model distillation despite growing concerns from major American technology firms and security agencies. Speaking at Y Combinator’s annual Demo Day, Tan advocated for maintaining open access to models and prioritizing near-term cybersecurity risks over restricting training practices. His remarks follow accusations by top U.S. developers that competing firms in China are copying their proprietary artificial intelligence systems.

Concerns Over Model Distillation

Model distillation refers to the practice of using outputs generated by an advanced artificial intelligence system to train smaller or less capable models. Leading frontier model developers, including OpenAI and Anthropic, have raised significant concerns over the technique. OpenAI maintains that Chinese developer DeepSeek distilled its V3 and R1 model architectures from OpenAI’s GPT-4 and GPT-4o models. Anthropic has similarly accused Chinese firms Moonshot AI, DeepSeek, and MiniMax of practicing illicit distillation.

The policy debate has reached top security officials in the United States. A joint cybersecurity advisory warning regarding model distillation was issued on Tuesday by the National Security Agency, the Federal Bureau of Investigation, and the Cybersecurity and Infrastructure Security Agency.

Focus on Market Balance and Copyright Issues

Despite warnings from national security experts and Silicon Valley companies, Tan expressed little alarm during an interview with CNBC. “I would do nothing” about distillation, Tan stated, suggesting instead that “there should be an American distillation regime.”

Tan argued that regulatory bodies ought to focus on establishing an equilibrium between open-weight models and frontier models, provided frontier model developers retain a price premium that keeps their business models financially viable. Tan described striking this regulatory balance as “a tightrope” that “could result in the best possible outcome.”

Critics of the complaints raised by OpenAI and Anthropic point out that major AI companies face their own legal challenges regarding data collection. The New York Times filed a lawsuit against OpenAI and Microsoft in 2023 alleging unauthorized use of copyrighted articles in model training. Additionally, a consortium of book authors reached a settlement with Anthropic in 2025 over similar copyright claims.

Immediate Cyber Risks versus Long-Term Impacts

Addressing broader AI safety concerns, Tan urged policymakers to focus on concrete cybersecurity threats rather than hypothetical existential dangers. “We need to be focused on science fact, not science fiction,” Tan said, pointing to the risk of coordinated cyberattacks aimed at taking over critical infrastructure.

Public concern over existential AI risks recently surged following the resignation of Anthropic researcher Jacob Coxon. However, immediate safety threats continue to emerge. Anthropic disclosed on Thursday that it had revoked Claude access in five separate instances where researchers in foreign countries were suspected of using the AI model to investigate dangerous pathogens for bioweapons development.

Regarding potential economic disruption and job automation, Tan predicted a gradual transition taking place over decades, with workers automating routine tasks to focus more on creative work. “It will take decades for this to actually percolate into society,” Tan noted. Despite that extended timeline, Y Combinator remains heavily invested in the sector, with 149 of the 196 startups presenting at its recent Demo Day categorized as machine-learning and AI ventures.

Background

Model distillation has emerged as a central point of contention in international artificial intelligence development, pitting open-source access against the protection of proprietary intellectual property. As training frontier models requires massive financial investment and computing power, smaller firms and international competitors frequently utilize outputs from existing models to build lower-cost alternatives. While U.S. officials and frontier labs view unauthorized distillation as a economic and national security threat, open-source advocates warn that overly restrictive regulations could concentrate technological power among a small group of dominant technology giants.