From the Accountability Gap to Accountable Decision Ownership: A Leadership-Centred Governance Framework for Agentic AI

Abstract

As artificial intelligence shifts from an assistive tool to an actor that initiates and executes consequential decisions, organisations confront a widening responsibility gap: it becomes unclear who owns a recommendation, who is empowered to contest it, and who carries the risk when an outcome is wrong. Emerging institutional frameworks address this gap at the level of organisational principle, yet they stop short of specifying who, in particular, is answerable once a machine acts. We argue that the prevailing technical and compliance-oriented framing of AI governance is incomplete, and that governing agentic systems is usefully understood as a leadership discipline concerned with the deliberate allocation of decision rights to named individuals. Building on the information systems literature on decision rights and on the agency of intelligent artefacts, we develop a conceptual framework, Accountable Decision Ownership (ADO), that separates three core accountability roles (decision owner, independent challenger, and risk bearer), supported by an enabling technical owner, and ties graduated machine autonomy to the consequence and reversibility of decisions. We ground the role separation in precedents from regulated domains outside AI, extend the framework to multi-agent ecosystems and to organisational transitions, and illustrate its application through a credit-decisioning case. We advance four testable propositions intended to guide empirical work. The framework contributes a leadership-centred reconceptualisation of accountability for autonomous systems at the level of the individual decision, and offers executives a pre-deployment diagnostic for locating accountability before, rather than after, failure.Changes in v2: The technical owner is repositioned as an enabling role that supports three core accountability roles. The role separation is now grounded in precedents from banking supervision, risk management, internal audit, and commercial aviation. An illustrative credit-decisioning case has been added (Section 9), together with a summary of the ecosystem extensions (Table 2). The literature base has been expanded and the argument revised throughout.