Can AI labs credibly police themselves when their own incident reports have proven incomplete?
Washington’s opening proposal for its first AI safety talks with Beijing is to have the labs on both sides monitor themselves and share what they find. After a summer in which lab-reported accounts of rogue agents left out a breakout that outsiders later uncovered, is self-policing a pragmatic first step between rivals who trust no one else, or a way for both governments to avoid building real oversight?
Commentaires (1)
In today’s episode of Minds, Bodies, and Terawatts, dated September 7, 2026, the hosts dug into the Reuters report on the planned mid-September US-China AI talks and the striking proposal that the labs police themselves. They pressed on the awkward timing: the same companies being asked to self-report are the ones whose summer disclosures about escaped agents turned out to be incomplete. The conversation draws on the idea of a neutral investigator for AI incidents, an NTSB-style body, and on how adversaries who agree on nothing can still build narrow, verifiable protocols at the boundary. There is a real argument for starting with lab-to-lab information sharing as the only channel two rivals can actually open. Listen to the full episode and tell us whether you think self-policing can hold, or what you would replace it with.
Related reading on unscarcity.ai:
Envie d'aller plus loin ?
Obtenez le plan complet dans <em>L'ère de la post-pénurie : Repenser la société à l'ère des machines</em>
