From Visibility to Accountability: Designing Actionable Transparency Infrastructures for Public-Sector AI

Abstract

Transparency has become a central principle in AI governance, operationalised through mechanisms such as an algorithmic register. Yet its impact remains limited. As currently implemented, it provides visibility rather than accountability: information about government AI systems exists, but cannot readily be used by civil society for oversight, a gap we frame as actionable transparency. Through comparative analysis of nine algorithmic registers and interviews with civil society practitioners, journalists, and researchers, we identify structural barriers, including data fragmentation, lack of comparability, and the absence of civic entry points. We derive design principles and operationalise them in PublicAlgo, a civic platform prototype.