Squee451: halt shipping less monitorable models even if they appear more aligned

Squee451 · x · 2026-09-05

In the same thread, Squee451 states plainly that less monitorable models should not ship even if they appear more aligned, arguing the practice clearly raises risks massively. He speculates that if Anthropic and every other company agreed to this, OpenAI would follow — while admitting that may be too optimistic.

Related event: CoT Monitorability Decline Sparks Safety Debate Over Alignment vs Oversight(15 posts)→

Original post →

More from AGI Musings

AGI Musings channel →