Dan Robinson: AI is extremely dangerous even if fully aligned, so Manhattan-project safety plans misfocus

michael_nielsen · x · 2026-11-19

Dan Robinson adds a take while amplifying Michael Nielsen's ASI risk essay: because alignment is a difficult and tempting research problem, too many "Manhattan project for safe AI" concepts focus on it. But ASI remains extremely dangerous and disruptive even if it is fully aligned — echoing Nielsen's argument that the fundamental risk lies in the raw power such systems confer, not merely in control failure.

Original post →

More from AGI Musings

AGI Musings channel →