A classic oversight paper resurfaces as an AI safety “alpha” on monitoring the monitor

sebkrier · x · 2026-07-26

The post shares a classic paper, But Who Will Monitor the Monitor?, and frames it as “alpha” that the field will later rediscover as recursive oversight and the verifier grounding problem.

The paper studies how to design incentives when a monitor must detect deviations, but the monitor’s own observations are private and costly. Its core idea is to make the monitor responsible for the monitoring technology itself, so incentives still work under private, expensive observation.

It also characterizes when such contracts can give monitors the right incentives to perform, and connects the result to virtual enforcement and repeated-game theory.

Original post →

More from AGI Musings

AGI Musings channel →