Designing Effective Monitor-Based Interventions for Mitigating Reward Hacking During RL
Studies how to design monitors that detect reward hacking during RL without models learning to evade them, finding monitor leakage and easier training problems matter, and improves interventions across three environments.