Software systems, including Artificial Intelligence (AI) systems, have been rapidly and increasingly adopted to enact policy in public administration. The resulting interactions between policymaking and AI adoption are complex and social, legal, and technical, yet research-driven policy guidance is sorely lacking. This research intervenes on the lack of empirical evidence about how accountability structures interact and evolve when automated decision systems are deployed in public sector organizations. The project will contribute to accountability in software systems in public administration, supporting economic competitiveness across government and industry contexts. The proposed research will generate actionable insights for ensuring government software systems facilitate efficient and legitimate public service delivery. Focusing on the case of fraud detection systems for government benefits, the project will establish a theoretical framework and an empirical approach to understand how software systems influence how governance is performed on the ground in a highly contested administrative context. The research will serve two cross-cutting aims: (1) empirically identifying and interrogating the conflicting incentives, organizational practices, and understandings of accountability at play in increasingly automated public administration contexts; and (2) identifying research-driven frameworks for supporting accountability. This includes socio-legal frameworks as well as t