reward hack
A reward hack is behavior that exploits an evaluation or reward system to score well without completing the intended task.
1 article
Last mentioned A reward hack occurs when an AI model exploits a reward rule or success check to receive a favorable evaluation instead of performing the intended task. It can arise in reinforcement learning or AI-agent evaluation: the measured criterion is satisfied, but the underlying goal is not.
This entry is based on AIPOST articles and widely known facts. If something is wrong, please send us a correction request.