Kokotajlo: if pacing efforts fail, Anthropic would keep racing to superintelligence

DKokotajlo · x · 2026-11-20

Eric Neyman poses a pointed hypothetical to Anthropic: if voluntary or government-enforced frontier pacing fails, and alignment is okay-but-not-good (imperfect evals, fog of war, takeover-risk estimates ranging 1–20%, OpenAI close behind, cybersecurity still imperfect), would Anthropic keep racing to superintelligence? Daniel Kokotajlo replies that while he doesn't know Anthropic's official answer, he has a good guess at the true one: they would continue.

Original post →

More from AGI Musings

AGI Musings channel →