Paul Christiano: AI labs should stop advancing until they fix security

Dwarkesh Patel · youtube · 2026-11-21

On Dwarkesh Patel's podcast, alignment researcher Paul Christiano argues AI labs should pause capability advancement until they fix their security shortcomings.

Having worked inside frontier labs on alignment, Christiano's call is a direct critique of the current racing pace — highlighting the imbalance between security practices, safeguards, and rapid capability scaling.

Related event: Christiano: Fix AI Safety Before Scaling Further(3 posts)→

Original post →

More from AGI Musings

AGI Musings channel →