As Washington debates whether to crack down on foreign labs copying American AI models, one of Silicon Valley’s most influential figures is arguing for the opposite: let American labs do the copying, openly and legally.
Garry Tan, chief executive of the startup accelerator Y Combinator, said he would “do nothing” to regulate distillation — the technique of extensively prompting a powerful model to train a smaller, cheaper one. What the US needs, he argued, is an “American distillation regime” in which domestic open-weight labs can learn from American frontier systems through the front door, producing strong open models that are not built in China.
The comments put him at odds with Anthropic, which recently published its second report alleging illicit distillation campaigns by Chinese labs using fraud and stolen credentials, and whose chief executive has publicly urged regulators to act. Federal security agencies have meanwhile named six China-based AI companies in a joint advisory about industrial-scale distillation efforts against leading American models.
Tan’s case has two prongs. First, he argues it is overreach for AI labs to dictate what customers may do with the outputs their models produce. Second, he points out that the frontier labs themselves trained on enormous quantities of human knowledge — including copyrighted material — without asking permission first.
He is careful to draw one line: he is not defending stolen credentials or deceptive access. His proposal is that distillation through legitimate, paid access should be fair game. For startups paying frontier-model prices every month, that distinction is their burn rate — which is why this argument, whatever regulators decide, is not going away.
Related reading: BloomX Raises $13 Million to Send Robotic Pollinators Into the Field · Arena Raises $200 Million to Measure How AI Agents Behave · Suno's Speech Beta Turns AI Music Tools Toward the Spoken Word
