Why a 30-Day Model Release Delay May Not Work

beffjezos · x · 2026-07-15

Referencing Robin Hanson's proposal that "frontier labs should share models with standards bodies for evaluation up to 30 days before release," the author argues this mechanism essentially acts as a capability and risk audit on frontier models.

The author contends that technological and capital machines possess "computational irreducibility"—many impacts only become apparent post-deployment. Thus, regulation should target concrete, existing issues rather than attempting to predict all hypothetical risks. They also doubt a 30-day delay yields significant safety benefits, suggesting it might only slow progress and allow competitors to catch up.

Original post →

More from Safety

Safety channel →