Autonomous and AI-driven systems frequently treat refusal—the decision not to execute an action—as a failure condition, a denial of service, or an instance of excessive conservatism. This framing implicitly assumes that legitimate authority is expressed primarily through action, and that non-action represents degradation. As autonomous systems increasingly operate under conditions of uncertainty, degraded coordination, and ambiguous integrity, this assumption becomes insufficient. This work reframes refusal as a legitimacy-preserving enforcement act rather than an error or absence of behavior. It distinguishes authority from control and optimization, and separates legitimacy from credibility, showing how systems that continue to act without the capacity to contract authority risk silent authority decay. Refusal is presented as a first-class system behavior: an intentional boundary assertion that preserves governance when justification for action weakens. The paper situates refusal alongside silence and safe-mode as composable forms of restraint, forming a coherent vocabulary for governance-first autonomous system design. Rather than advocating inaction, it argues that the ability to refrain from action is a prerequisite for durable autonomy. Systems that cannot refuse, wait, or withdraw authority do not fail safely; they fail by continuing. This work contributes a conceptual framework for understanding refusal as enforcement, clarifying when and why restraint preserves legitimacy in autonomous systems operating beyond continuous supervision.
David Forbes (Sun,) studied this question.