Y Combinator CEO Garry Tan has a surprising stance on AI distillation, and it puts him at odds with some of the biggest names in American AI. In an interview with CNBC earlier this week, Tan said he would do nothing to stop distillation and even suggested the United States could benefit from its own distillation regime. His comments came just days after Anthropic released its second report alleging that Chinese labs are engaged in illicit distillation attacks.
What Garry Tan Said About Distillation
When asked about Chinese AI labs using distillation techniques to extract knowledge from frontier model makers, Tan did not call for a crackdown. Instead, he told CNBC, “I would do nothing.” He then went further, saying, “We could argue that there should be an American distillation regime.”
He elaborated to TechCrunch that this means he wants smaller, American open-weight AI labs to use the same kind of training techniques on American frontier AI labs. The goal is to give the United States a more robust set of open-weight options that are not Chinese.
To be clear, Tan is not advocating for American AI labs to use stolen credentials to distill. He wants them to be free to come in the front door. His argument is twofold, and both parts deserve a closer look.
How Distillation Works
Distillation is when a model maker extensively prompts another model in order to learn how it works and reasons. It is commonly, and legitimately, used by AI labs to help train new models. The technique itself is not inherently illegal or even unusual.
The controversy arises when distillation is done without permission, using hidden identities, fraud, or stolen credentials. Anthropic this week released its second report alleging that Chinese labs are engaged in what it calls “illicit distillation attacks.” The report claims these labs hide their identities to distill without permission.
Anthropic CEO Dario Amodei had previously publicly called on U.S. regulators to crack down on distillation. Tan’s position stands in notable contrast to that request, especially given his role leading one of Silicon Valley’s most prestigious and prolific startup accelerators.
Tan’s Two-Part Argument
Tan’s first argument is about control. He feels it is an overreach for AI labs to dictate what their customers can do with the information their models share with them. “Controlling what users and customers do with API calls to closed weight models feels constraining,” he told TechCrunch.
He added that there is a role government can play here to normalize the fact that access to intelligence trained on broad public access data should itself also be more a form of a public good than something locked away behind restrictive terms of service.
His second argument centers on consistency. He notes that the proprietary AI labs did not ask permission when they vacuumed up as much human knowledge as they could to train their models. They famously ingested plenty of copyrighted material without the permission of those intellectual property holders. For Tan, it is hard to justify strict control over model outputs when the inputs were gathered so freely.
The Balance Between Open-Weight and Frontier Labs
Tan does not want frontier labs to disappear. He wants to see a balance between open-weight AI labs and frontier labs. “They are at the frontier and driving it forward. We want that to be fundable, and be a great business model ongoing,” he told CNBC. “You want open weight models to give people freedom and access.”
That dual vision is central to his thinking. Frontier labs push the boundaries of what AI can do. Open-weight models distribute that capability widely. Tan believes both can coexist, and that American open-weight labs should be allowed to compete without being treated as bad actors for using standard training techniques.
The Doomer Scenario
For Tan, the true AI doomer scenario is not distillation. It is concentration of power. “The nightmare scenario, the doomer scenario for AI is that there’s just one company,” he said. “It has the best access to capital. It has the best AI researchers. It runs away with it and suddenly there’s one company that’s monolithic. And that would be bad.”
This concern shapes his entire position. If distillation is heavily restricted, only the largest and most well-funded labs can afford to train frontier models from scratch. Smaller open-weight labs would be shut out. The result could be the very monopoly Tan fears.
Tan, who is himself such an avid AI user that he once described himself as having cyber psychosis, wants to see a healthy ecosystem rather than a single dominant player.
Why the Debate Matters
The distillation debate sits at the intersection of several unresolved questions in AI policy. Who owns the knowledge that comes out of a model? Can a company restrict how its API is used? Should the government intervene to protect frontier labs from competition, or to promote open access?
Anthropic’s allegations against Chinese labs raise legitimate security and fraud concerns. Tan’s position does not deny those concerns. He specifically opposes stolen credentials and hidden identities. His argument is that legitimate, front-door distillation by American labs should not be lumped in with illicit attacks.
That distinction may prove important as regulators consider how to respond. A blanket crackdown on distillation could harm the open-weight ecosystem that many developers and researchers rely on. A targeted approach focused on fraud and unauthorized access might address the security concerns without stifling competition.
What Happens Next
Tan’s comments add a prominent voice to a debate that is still unfolding. Anthropic has now released two reports on the issue and its CEO has called for regulatory action. Tan, leading one of the most influential startup accelerators in the world, is pushing back.
The outcome will shape how open-weight AI develops in the United States. If Tan’s view prevails, American open-weight labs could gain more freedom to train on frontier models. If Anthropic’s view prevails, restrictions could tighten. Either way, the question of how AI labs learn from each other is not going away.
For now, Tan’s message is clear: let American labs distill, and worry less about the technique itself than about the concentration of power that restrictive policies could create.
