Garry Tan calls for a U.S. open-weight distillation regime, arguing American labs should freely distill frontier models rather than regulators banning the practice.
September 11, 2026
brightray analysis
Summary
Y Combinator's CEO argues that rather than cracking down on distillation—where one model is trained by extensively querying another—the U.S. should encourage domestic open-weight labs to distill American frontier models. His case rests on two points: frontier labs lack moral standing to restrict API-derived learning given their own unconsented use of copyrighted training data, and AI power concentrated in a single proprietary provider is the true catastrophic risk.
Why it matters
- Tan proposes a formal 'American distillation regime' giving open-weight U.S. labs the right to distill frontier models—directly contradicting Dario Amodei's call for regulatory crackdowns.
- He argues frontier labs forfeited their moral authority to restrict distillation by training on copyrighted content without permission, framing API-derived learning as a near-public good.
- His primary AI risk framing is monopolization—one company owning all frontier capability—not open-weight proliferation, inverting the standard safety narrative.
Community notes—
No notes yet — be the first.
See every signal in the Feed