Y Combinator CEO Garry Tan Urges Regulators to Protect Open-Weight AI
Key Takeaways
- •Garry Tan called on regulators on September 10 to leave open-weight AI models largely unrestricted, framing frontier models as a public good built on publicly available human knowledge.
- •His remarks followed a joint NSA, FBI, and CISA advisory that flagged alleged distillation of OpenAI and Anthropic model outputs by Chinese companies including DeepSeek, Moonshot AI, and MiniMax.
- •Tan said he would do nothing about these Chinese firms, portraying distillation—training smaller models to mimic larger ones—as an inevitable consequence of building AI on public data.
- •He argued that restrictions on distillation could damage the broader AI ecosystem more than they would protect any company's intellectual property, suggesting labs use pricing strategies to maintain their edge.
- •Of the 196 startups presenting at Y Combinator's Demo Day, 149—roughly 76%—were AI or machine learning companies, and Tan backs open-source efforts through projects like GBrain and GStack.
Garry Tan, president and CEO of Y Combinator, used the accelerator’s Demo Day on September 10 to urge regulators to leave open-weight artificial intelligence models largely unrestricted. Tan argued that frontier AI models, built on the foundation of publicly available human knowledge, should serve as “a form of public good.”
His remarks came days after a joint advisory from the NSA, FBI, and CISA raised concerns about alleged large-scale model-distillation efforts by Chinese companies. Tan said he would “do nothing” about Chinese firms such as DeepSeek, Moonshot AI, and MiniMax, which have reportedly distilled outputs from models developed by OpenAI and Anthropic.
Rather than describing distillation as an existential threat, Tan presented it as an inevitable consequence of developing powerful AI systems using publicly available data. Distillation involves training smaller models to reproduce the behavior of larger models.
Tan argues for open access
Tan called on regulators to preserve “freedom and access” in AI development. In his view, restrictions on distillation could damage the broader ecosystem more than they would protect the intellectual property of any individual company.
He offered a structural argument for limiting regulation: If frontier AI laboratories want to protect their competitive advantage, they can rely on pricing strategies instead of regulatory barriers.
Y Combinator’s AI-heavy cohort
The composition of Y Combinator’s latest batch reflects the importance Tan places on AI development. Of the 196 startups that presented at Demo Day, 149 were machine learning or AI companies, representing roughly 76% of the cohort.
Tan has also supported his position through projects such as GBrain and GStack, which promote open-source AI tools. His call for “greater aggressiveness in pursuing open-source models” indicates that his support for open development extends beyond a general philosophical position and is part of his broader approach at Y Combinator.
National security concerns
The joint NSA-FBI-CISA advisory identified specific Chinese firms, elevating what had previously been an open industry secret into an official national security concern. By rejecting what he characterized as doomsday concerns about AI advancement, Tan redirected attention toward cybersecurity, infrastructure protection, and maintaining the United States’ innovation edge through openness rather than restriction.
The competing positions highlight the policy question at the center of the debate: whether the benefits of wider access to AI models outweigh concerns about model outputs, intellectual property, and national security. Tan’s comments favor access and market-based protection, while the advisory underscores that regulators and security agencies are also examining how advanced models are reproduced and used.
Source: CryptoBriefing
