Responsible AI Risk Governance in Public Service Organizations—Insights from a Meta-Synthesis

The use of AI in public service provision can reshape citizen-state interactions, introducing new risks and challenges to ensure safe, inclusive, and just applications that do not discriminate against vulnerable citizens. This study examines how public service organizations govern these risks. Drawing on a meta-synthesis of AI risk governance scholarship (n=56), we synthesize the empirical evidence from a value-based risk governance perspective. We identify four governance challenges: (1) establishing substantive pre-assessment before adoption; (2) enabling informed, plural, and revisable appraisal; (3) ensuring meaningful human oversight during implementation; and (4) preserving organizational ownership in ensuring that monitoring and feedback are connected to correction and learning. For each challenge, the synthesis identifies failure patterns and proposes solutions on how to avoid them. Integrating these findings, we develop a novel evidence-based AI risk governance framework for public organizations. Advancing the discourse on technology risk governance and administrative burden in the age of AI, this study provides an organizational explanation for AI risk governance success and failure, offering insights for theory and practice.