Monitoring Reasoning Models for Misbehavior
Monitoring Reasoning Models for Misbehavior
Why it matters
Optimizing the CoT directly makes reward hacking evolve into a hidden form — the basis for the norm of not putting optimization pressure on the CoT
Source — Primary sources first; secondary reports are labelled “Press”
Paperarxiv.orghttps://arxiv.org/abs/2503.11926Added in Week 35, 2026 · See that week