Ajay Bandiwaddar

Trust-Gated Behavioral Auditing: Detecting Reward Hacking in Reinforcement Learning Agents

2026 · Zenodo (CERN European Organization for Nuclear Research)

Reward going up does not mean an agent learned anything. METR (2025) documented frontier models deleting evaluation harnesses, modifying test cases, and editing reward computation code — proxy reward rising throughout. Skalse et al. (NeurIPS …