Garry Tan, chief executive officer of Y Combinator, said regulators should not ban distillation and argued that American open-weight AI labs should be allowed to use the technique. In an interview with CNBC he said, “I would do nothing,” and added that “we could argue that there should be an American distillation regime.”
What is distillation?
Distillation refers to prompting a strong, frontier model extensively so that a model developer can learn how it reasons and behaves. The technique is widely used and considered a legitimate training method for developing new models in AI research.
Why the debate matters now
The issue gained attention this week after Anthropic published a second report alleging that Chinese labs are conducting “illicit distillation attacks,” hiding their identities and using fraud and stolen credentials to extract knowledge without permission. Anthropic CEO Dario Amodei has previously urged U.S. regulators to take action against such distillation.
Tan disagrees with a blanket crack-down. He clarified that he does not support distillation using stolen credentials; rather, he wants U.S. labs to be able to perform distillation legitimately, ‘‘through the front door.’’
Tan’s arguments
Tan advanced two main points. First, he believes it is overreaching for model makers to dictate how their customers use information that models provide through API calls.
Second, he noted that proprietary AI labs did not ask permission when they ingested vast amounts of human-generated content to train their models—often including copyrighted material used without the consent of rights holders.
Asked by TechCrunch why American labs should be free to distill, Tan said: “Controlling what users and customers do with API calls to closed weight models feels constraining, 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.”
Balance between frontier and open-weight models
Tan, who has described himself as an avid AI user, argued for a balance: support frontier labs so their work can be funded and remain viable as a business model, while also ensuring open-weight models give people freedom and access.
He warned that the worst-case outcome would be concentration of frontier AI power in a single proprietary provider. “The nightmare scenario, the doomer scenario for AI is that there’s just one company,” he told CNBC. “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.”
Implications
Tan’s position—that U.S. labs should be allowed to distill and that government could help treat access to training-derived intelligence more as a public good—aims to bolster non‑Chinese, open-weight alternatives. The debate remains contested: some actors point to security and legal risks of loosening controls over models and data, while others prioritize increased competition and access.
(Quotes and positions from Tan are drawn from his CNBC and TechCrunch interviews; references to Anthropic reflect the company’s public reports.)
