What is reward hacking in Q-Learning systems?

Reward hacking occurs when the agent exploits flaws in reward design.

The agent finds unintended strategies that maximize reward without achieving the true goal. This happens due to misaligned reward functions and insufficient constraints.