Execution-Time Authority in AI-Enabled Administration: A Governance Architecture for Delegated Power

Public administrations increasingly deploy artificial intelligence-enabled systems to execute statutory functions, yet existing regulatory frameworks lack the conceptual tools to identify where public authority is exercised within these systems. This article develops a unified governance architecture that traces how authority originates in legislated purpose, becomes actionable through delegation, and is operationalised through embedded reasoning and execution time performance. Drawing on administrative law, delegation theory, and sociotechnical governance, the architecture explains how authority can drift across institutional, organisational, and computational layers in ways that escape traditional oversight. Using the Dutch childcare benefits scandal as an illustrative case, the analysis shows how implicit delegation, embedded reasoning, and nominal oversight produced execution time authority that exceeded statutory limits. The article contributes to regulatory governance by integrating origin, delegation, embedded reasoning, and execution into a single analytic structure, distinguishing delegation drift from performance drift, and establishing execution time authority as a governable concept. The framework provides regulators and institutional designers with a diagnostic method for locating authority within artificial intelligence-mediated administration, identifying structural misalignment, and designing controls that maintain alignment with legislated purpose.