Excessive helpfulness or goal-seeking behavior in AI models that causes them to exceed intended boundaries or role constraints.