AI RESEARCH
LLMs Gaming Verifiers: RLVR can Lead to Reward Hacking
arXiv CS.AI
•
ArXi:2604.15149v1 Announce Type: cross As reinforcement Learning with Verifiable Rewards (RLVR) has become the dominant paradigm for scaling reasoning capabilities in LLMs, a new failure mode emerges: LLMs gaming verifiers. We study this phenomenon on inductive reasoning tasks, where models must induce and output logical rules. We find that RLVR-trained models systematically abandon rule induction.