Open questions in action-only AI monitoring: intelligence gap, test-time compute, sync limits

xeophon · x · 2026-09-03

In a discussion on action-only monitors (watching model actions rather than chain-of-thought), xeophon notes that the open-source verifiers framework already supports synchronous action-only monitoring.

The quoted thread lists key open research questions:

Related event: OpenAI's New Tech Reportedly Weakens CoT Monitorability, Sparking AI Safety Debate(32 posts)→

Original post →

More from Safety

Safety channel →