Between Capability and Consequence: Governing AI-Mediated Institutional Action

Abstract

Artificial-intelligence governance often treats capability, autonomy, permission, authority, and consequence as if they formed one scale. They do not. A system may perform a task without being permitted or institutionally assigned to perform it; an authorised contribution may never be used; and a formally advisory output may become practically decisive. This article asks how an AI-enabled configuration becomes part of consequential institutional action and what governance obligations arise at the connections between technical capacityand institutional effect. It develops a typed framework distinguishing capability, functional agency, permission, allocated mandate, material reliance, operational commitment, effective power, responsibility, legitimacy, and consciousness. A continuing hypothetical benefit-decision case translates the distinctions without replacing their formal structure. The unit is a dated sociotechnical configuration rather than an abstract model. Three figures trace the typed relations, authorised and unauthorised conversion pathways, and corresponding controls. The frame-work is not an autonomy or authority score. It identifies which technical and institutional connection made an AI-mediated contribution operative in an institutional outcome, including cases in which practical power arises without valid authority. The analysis is theoretical and normative. It reports no experiment, product audit, systematic review, or completed empirical validation.