Algorithmic Authority in Practice: Generative AI, Decision-Making, and Accountability in Public Administration

Generative AI (GenAI) is reshaping public sector decision-making, yet empirical understanding of how it reconfigures governance remains limited. Drawing on 24 semi-structured interviews with public officials in Kenya, this interpretive qualitative study examines how GenAI reshapes decision-making authority, accountability, and trust in human-centred governance. The findings identify three interrelated mechanisms: the concentration of epistemic authority in generative outputs, strongest among officials lacking the technical or situated knowledge to interrogate them; a structured pattern of default compliance and selective override, where organisational norms place the burden of justification on deviation rather than compliance; and a diffusion of accountability through system-based attribution for co-produced outputs, reversible where organisations maintained clear review and documentation. GenAI’s fluent, narrative outputs, unlike the scores and flags of earlier systems, appear to lend a persuasive authority we interpret as theoretically significant rather than formally demonstrated. Officials perceived that resulting opacity and diffused responsibility could strain institutional trust even where performance improves, reflecting their accounts rather than direct evidence from citizens. The study contributes a process-oriented, GenAI-specific analytical framework, empirically grounds hybrid discretion in organisational practice, extends emerging accounts of GenAI-specific accountability entanglement, and offers a theoretically informed account of officials’ perceptions of GenAI’s implications for institutional trust.