From Technical Capability to Institutional Authority:

Abstract

Artificial intelligence governance often treats permission as though it completes the problem of authority. Yet a system may be technically capable of acting, connected to operational resources, and supported by an earlier approval while the proposed action is no longer institutionally legitimate when execution occurs. This article distinguishes technical capability, authorization, legitimacy, and execution admissibility. It argues that consequential AI-enabled action requires more than access or prior approval: it requires valid institutional authority, identifiable decision ownership, sufficient and current evidence, applicable policy conditions, and a genuine capacity for human intervention. The article names the interval between prior authorization and present execution the execution gap. It then develops the execution gate as an institutional mechanism through which an action may be allowed, held, denied, or escalated before technical capacity becomes real-world consequence. Human responsibility, on this account, must be constituted before action, not reconstructed only after harm.