Garry Tan, CEO of Y Combinator, wants distillation left alone - as long as it happens in the United States. American open-weight labs should be free to distill American frontier models, and regulators should, in his words, “do nothing”. His term for it: an American distillation regime.
His argument
Frontier models were trained on publicly available human knowledge, so access to capable AI is a form of public good. Telling people what they may do with the answers their own API calls produced strikes him as overreach.
There is a jab attached: the proprietary labs did not ask anyone’s permission either when they poured copyrighted material into training.
What worries Tan is not the open model but concentration. The real doomer scenario, he says, is one company owning the market. At the same time he wants frontier research to stay a viable business.
The other side
That sits almost exactly opposite Anthropic. Dario Amodei called for a regulatory crackdown after Anthropic accused Chinese labs of illicit distillation attacks involving fraud and stolen credentials. This week’s threat report names Moonshot and DeepSeek.
And in today’s essay, Amodei lists cracking down on unauthorized distillation as one of three measures democracies need in order to hold their lead.
Why this one stuck with me
Both men use the same word for two different things. Tan means American companies distilling through the API. Anthropic means actors evading access controls and faking accounts.
A law that hits the second and leaves the first alone would have to draw that line cleanly. Nobody has shown how yet - which is exactly what makes Tan’s “do nothing” so comfortable a position.
Sources: