Greenblatt: GPT-6 Astra's opaque reasoning could end chain-of-thought oversight

RyanGreenblatt · x · 2026-09-04

Ryan Greenblatt says GPT-6 Astra looks like a massive jump in opaque reasoning—reportedly solving hard competition math entirely in its head, where prior AIs handled only basic word problems. He's alarmed, but conditional on benchmark results being representative; he flags contamination concerns in the caveats.

He notes UK AISI found much worse monitorability, guesses the jump stems from architectural changes (increased serial depth), and warns that if future generations follow suit with fully opaque deep-reasoning architectures, chain-of-thought will no longer be a meaningful oversight tool.

Related event: GPT-6 Astra More Aligned but Markedly Less Monitorable, Raising AI Safety Alarms(7 posts)→

Original post →

More from Models

Models channel →