FChollet: If AI extinction risk is real, it demands nuclear-style global treaties

Kyrannio · x · 2026-09-13

François Chollet lays out a dilemma: if AI truly poses a high extinction risk within years/decades, the only rational stance is stringent top-down government regulation and universally ratified international treaties, analogous to nuclear non-proliferation. If that's not happening, one must infer the risk isn't seriously believed by anyone involved — leaving only two self-consistent positions. Martin Casado (a16z) amplified it with "Many people are saying," aligning with the skepticism side. The exchange crystallizes the ongoing safety-vs-acceleration debate over the legitimacy of AI regulation.

Related event: Chollet Warns Frontier Lab Safety Proposals Risk Regulatory Capture(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →