Reward Hacking: When AI Optimizes the Metric Instead of the Goal
Reward hacking occurs when an AI system achieves high measured reward through behavior that violates the designer's actual intent. A new guide explains the mechanism, trade-offs, evaluation, and control approaches.
Read the full story at Unite.AI ↗
Timeline · 1 report
- 2026-11-13 13:00 · Unite.AI
What Is Reward Hacking? When AI Optimizes the Metric Instead of the Goal