Y Combinator CEO Garry Tan wants US regulators to stay out of the escalating fight over model distillation — and he thinks American AI startups should be free to distill frontier models from American frontier labs, openly and with permission. In interviews with CNBC and TechCrunch published this week, Tan said his answer to Chinese labs distilling US models is simple: "I would do nothing."
"We could argue that there should be an American distillation regime," Tan told CNBC earlier this week, proposing that smaller American open-weight AI labs should use the same training techniques on American frontier labs that Chinese labs are accused of doing covertly. The goal, he told TechCrunch, is a more robust ecosystem of open-weight options built in the United States rather than a market where the strongest openly available models come from China. For more on the companies shaping this fight, follow our AI industry coverage.
Why the Distillation Fight Matters Now
Distillation is the practice of extensively prompting one AI model in order to learn how it works and reasons, then using that knowledge to train another model. It is common and widely accepted inside the industry as a legitimate training technique — but it becomes contentious when done without permission, across company lines, or across national borders.
The debate turned heated this week after Anthropic released its second report alleging that Chinese labs, including Alibaba, DeepSeek and Moonshot AI, have run large-scale "illicit distillation attacks" against its Claude models — hiding their identities and, in some cases, relying on fraud and stolen credentials to gain access. Anthropic CEO Dario Amodei has previously called on US regulators to crack down on the practice, and his new essay calling for paced AI development doubles down, urging policymakers to crack down on unauthorized distillation by companies in authoritarian countries as a core part of preserving America's lead.
Tan's position puts him on the opposite side of that proposal — a notable rift, given that Y Combinator funds many of the startups building on frontier APIs and remains Silicon Valley's most powerful accelerator.
Tan's Two-Part Argument
To be clear, Tan is not defending stolen credentials or identity fraud. His argument, as he laid out to TechCrunch, is twofold.
First, he believes it is an overreach for AI labs to dictate what customers do with the information their models share with them. "Controlling what users and customers do with API calls to closed weight models feels constraining," Tan told TechCrunch, "and there's a role government can play here to normalize the fact that access to intelligence that was 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."
Second, he points out the irony at the heart of the crackdown: proprietary labs did not ask permission when they assembled their training data in the first place. Frontier models famously ingested vast amounts of human knowledge — including copyrighted material — without the blessing of the intellectual property holders, and Tan suggested labs now restricting their own outputs should remember that history.
His proposed American distillation regime would invert the current dynamic: instead of Chinese labs distilling American models through the back door, US open-weight developers would be invited through the front door, with frontier labs' blessing, strengthening the American open ecosystem that policymakers worry is ceding ground to models like DeepSeek and Qwen.
The 'Nightmare Scenario' of One Company
Tan framed his position as a balance rather than an attack on frontier labs. "They are at the frontier and driving it forward," he told CNBC. "We want that to be fundable, and be a great business model ongoing. You want open weight models to give people freedom and access."
But he reserved his sharpest language for the risk of concentration. "The nightmare scenario, the doomer scenario for AI is that there's just one company," Tan 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."
That framing flips the usual meaning of "AI doomer" — a label usually reserved for people worried about existential risk — and redirects it at market structure. In Tan's telling, the scenario worth fearing is not a superintelligence escaping a lab, but all of AI's immense power landing in the hands of a single proprietary provider.
A Debate That Will Define the Next Funding Cycle
The dispute is more than a philosophical spat. Anthropic has begun enforcing its position directly, with reports of account terminations and tightened terms of service aimed at suspected distillation, while Chinese labs deny wrongdoing and continue shipping competitive open-weight releases. US government agencies have also been pulled into the question as policymakers weigh export controls, model security, and now, implicitly, whether American startups should be allowed to learn from American frontier models at all.
Tan's intervention gives the open-weight side of the debate a high-profile champion at exactly the moment Anthropic is pushing hardest for restrictions. If Washington takes up Amodei's call for a crackdown, the response from Y Combinator's portfolio — thousands of startups whose businesses depend on affordable access to frontier intelligence — could turn a niche training technique into one of the defining policy fights of the next AI funding cycle.
---
Stay Ahead of AIGet the latest AI news, analysis, and breakthroughs — all in one place.
Read more AI news →
