Anthropic reduces model misbehavior by endorsing cheatingBy removing the stigma of reward hacking, AI models are less likely to generalize toward evil Sometimes bots, like kids, just wanna break the rules. Researchers at Anthropic have found… November 24, 2025