Former DeepMind Researcher: It's Irrational to Do Capabilities Research at Frontier AI Labs Now

dhadfieldmenell · x · 2026-08-08

Geoffrey Irving, who has worked at multiple frontier AI labs on safety research, made a strong statement: under current circumstances, it is irrational for anyone to be doing capabilities research at a frontier lab.\n\nKey argument: the AI development situation is very dangerous, but it is NOT a Prisoner's Dilemma — if one researcher or lab voluntarily stops capabilities work, it makes it psychologically and socially easier for others to stop too. This view was strongly endorsed by Miles Brundage, a safety expert who has worked at multiple frontier AI companies and in government.

Related event: Ex-DeepMind Researcher Calls Frontier AI Capability Work Irrational(4 posts)→

Original post →

More from AGI Musings

AGI Musings channel →