Garry Tan, the CEO of Y Combinator, has thrown his weight behind a position that could reshape the balance of power in artificial intelligence: U.S. open-weight AI labs should be allowed to distill frontier models, too.
His argument is deceptively simple. Frontier AI models were themselves trained on publicly available human knowledge — books, websites, research papers, code repositories. If the inputs were public, Tan suggests, then access to capable AI should be treated as "a form of public good."
That single framing puts him at the center of one of the most consequential debates in tech right now: who gets to build with the most powerful AI systems, and under what rules.
What Tan Actually Said — and What Distillation Means
Distillation, in AI terms, is the process of training a smaller, more efficient model to mimic the behavior of a larger, more capable one. It's how many open-weight models achieve surprising performance without the massive compute budgets of frontier labs.
Tan's position is that if frontier models learned from public human knowledge, then restricting distillation to only a handful of well-resourced companies runs counter to the spirit of open access. He frames capable AI not as a luxury product but as infrastructure that more builders should be able to use.
Why This Argument Lands Differently in 2025
The U.S. AI landscape is increasingly split between closed frontier labs — which guard their model weights and terms of service tightly — and open-weight labs that publish models for anyone to download and modify.
Distillation sits awkwardly between them. It's not outright copying, but it's also not independent development. For open-weight labs, it's often the fastest path to competitive performance. For frontier labs, it can feel like their research investment is being repackaged and redistributed.
Tan's intervention matters because Y Combinator has funded a significant number of AI startups, many of which depend on access to capable models. His view reflects a segment of the ecosystem that sees open access as a growth engine, not a threat.
The Public Good Framing — and Its Limits
Calling AI access a "public good" is a strong rhetorical move. In economics, a public good is something that is non-excludable and non-rivalrous — like clean air or national defense. AI models don't fit that definition neatly, since they require enormous compute, energy, and talent to create and run.
Still, Tan's underlying point resonates: the knowledge that trained these models was collectively produced. The question is whether that justifies a right of access to the models themselves, or simply to the knowledge they were built from.
That distinction is where the legal and ethical debate gets genuinely difficult.
Who Benefits — and Who Pushes Back
For startups, researchers, and smaller labs, looser distillation rules could mean faster experimentation and lower costs. For frontier labs, it could mean diluted competitive advantage and harder-to-enforce terms of service.
Critics of open distillation argue that it can strip away safety guardrails, make misuse harder to trace, and undermine the incentive to invest billions in frontier research. Supporters counter that openness has historically driven innovation and that concentrated AI power carries its own risks.
Neither side has a clean answer yet. What's clear is that the status quo — where a few labs control the most capable models — is being questioned from within the industry itself.
Confirmed Facts vs What Remains Unclear
Confirmed: Garry Tan has publicly argued that U.S. open-weight AI labs should be allowed to distill frontier models, and has framed capable AI access as a form of public good.
Unclear: Whether this position will translate into formal policy proposals, regulatory advocacy, or changes in how Y Combinator-backed companies operate. No specific legislative or corporate action has been confirmed.
Speculation: Any predictions about how frontier labs or regulators will respond remain speculative at this stage.
The Bigger Pattern: AI's Open vs Closed Fight Is Heating Up
Tan's comments are part of a broader tension that has been building for years. Open-weight models have closed the performance gap faster than many expected. Governments are watching. Investors are repositioning. And the question of who owns — or should own — AI capability is no longer just a technical debate.
It's now a policy debate, an economic debate, and increasingly a cultural one.
What This Means for Builders and Researchers
For now, nothing has changed legally. Distillation rules, terms of service, and export controls remain as they are. But the conversation Tan is pushing could influence how those rules evolve.
Builders should watch for signals from regulators, frontier labs' terms of service updates, and any movement from open-weight advocacy groups. The direction of U.S. AI policy in the coming months will matter more than any single statement.
Future Outlook
If Tan's view gains traction, the next few years could see a more permissive environment for open-weight development in the U.S. If it doesn't, the gap between frontier and open labs may widen further.
Either way, the debate he's surfaced — about public knowledge, public good, and who gets to build — isn't going away.
Our Take
Tan's argument is provocative because it reframes distillation not as a loophole but as a logical extension of how AI was built in the first place. Whether regulators and frontier labs accept that framing is another matter. But by putting the "public good" language into the conversation, he's shifted the terms of debate — and that alone makes this worth watching.
Frequently Asked Questions
What does Garry Tan want open-weight AI labs to be allowed to do?
He wants them to be permitted to distill frontier models — training smaller models to mimic larger ones — arguing that capable AI access should be treated as a form of public good.
What is model distillation in AI?
Distillation is a technique where a smaller model is trained to replicate the behavior of a larger, more capable model, often achieving strong performance at lower cost.
Why is this controversial?
Frontier labs invest heavily in their models and often restrict how they can be used. Distillation by open-weight labs can be seen as undermining that investment or bypassing safety controls.
Has any policy changed because of Tan's statement?
No. Tan's statement is a public position, not a formal proposal. No regulatory or legal change has been confirmed.
