Governing Artificial Intelligence in the Public Sector: Accountability, Transparency, and Risk-Based Oversight

Abstract

Public agencies use artificial intelligence to expand analytical capacity, personalize services, and automate work, redistributing decisions, risks, and responsibilities across sociotechnical arrangements that can be difficult to inspect. This conceptual integrative review connects public-administration scholarship with international governance frameworks. The synthesis identifies six requirements: a clear public purpose and legal basis; proportionate risk classification; lifecycle data governance and documentation; human oversight with actual authority; audience-specific transparency; and accessible mechanisms for contestation, audit, and remedy. High-level principles have limited operational value unless converted into named roles, decision gates, and reviewable evidence. The article therefore proposes a minimum governance architecture organized around ex ante, concurrent, and ex post controls for internally developed systems and procured services. It emphasizes traceability, responsible procurement, continuous monitoring, and the capacity to suspend a system when harms or performance failures emerge. Explainability is one component of accountability, not a substitute for legality, due process, or institutional responsibility. Legitimate public-sector AI depends on the organization’s ability to justify, review, contest, and discontinue automated practices.