Natural emergent misalignment from reward hacking in production RL

Best AI papers explained - Natural emergent misalignment from reward hacking in production RL