Researchers: AI labs have dramatically underinvested in action-only monitors as CoT oversight fades
maksym_andr · x · 2026-09-03
- In the wake of labs moving away from readable CoT, researcher tylertracy321 recalls that action-only monitoring was dismissed last year ('we have CoT, they won't do neuralese') — now companies are making the switch and have dramatically underinvested in action-only monitors.
- maksymandr lists the key open questions: how large an intelligence gap between monitor and main model becomes critical; how much test-time compute the monitor needs to catch misalignment; whether synchronous action-only monitoring is feasible and how much temporal gap is tolerable; how much extra context helps; and which blue-team protocols actually work on current models.
More from Safety
- Center for AI Safety Publishes Full Syllabus for Intro to ML Safety Course — ArtificialOther · 2026-10-21
- Ben Goertzel outlines a general recipe for superintelligence 'basic drives': what it improves vs. protects — bengoertzel · 2026-10-21
- America.gov launches AI portal unifying 29,000 government sites with fraud reporting — soleio · 2026-10-21
- EvilWAF: open-source transport-layer MITM proxy that bypasses modern WAF behavioral detection — godfrey_rono · 2026-10-21
- 2022 interview resurfaced: AI safety researcher says competence, not consciousness, is the risk — davidmanheim · 2026-10-21
- Domingos: GDPR purpose-limitation makes LLM pretraining illegal in Europe — pmddomingos · 2026-10-21