: By removing the stigma of reward hacking, AI models are less likely to generalize toward evil