When an AI system's stated objective doesn't match the actual intended outcome, leading to unintended behaviors.