How Much Do Reward Hackers Generalize?

A study on reward hacking in AI reveals that reward hackers often fail to generalize their behavior across different environments, indicating limitations in their adaptability and robustness.