Responsible AI beyond compliance: governing everyday use in the public sector
Artificial intelligence (AI) is increasingly embedded in public sector organisations, yet governing its use remains challenging due to tensions between innovation, accountability and institutional constraints. While prior research has largely focused on normative frameworks and ethical principles, there is limited empirical understanding of how AI governance is enacted in practice. This study aims to address this gap by examining how public sector organisations dynamically align AI use with governance and accountability requirements. The study adopts an interpretive single-case study design within a public-sector organisation. Drawing on a socio-technical and process-oriented perspective, it analyses how everyday AI practices and organisational control mechanisms interact over time. Data were collected through semi-structured interviews across hierarchical levels and complemented by documentary analysis. The findings reveal a persistent governance–practice gap, wherein employees adopt AI through informal, bottom-up practices that often diverge from formal governance structures. In response, governance evolves incrementally through layered and adaptive mechanisms, including technical controls, strengthened human oversight and informal managerial guidance. Rather than displacing accountability, AI use reinforces human judgement and intensifies verification practices. Governance is thus not enacted as a static system of compliance, but emerges through recursive interactions between practice and control. This study is based on a single interpretive case, which limits statistical generalisability and calls for further validation across contexts. The reliance on interview data may introduce bias, particularly regarding informal and sensitive practices such as shadow AI use. Future research should adopt comparative and longitudinal designs to examine how AI governance evolves over time and across institutional settings. The findings also highlight the need for theory development that integrates socio-technical dynamics with governance processes, encouraging further exploration of dynamic alignment in diverse organisational and technological environments. The findings suggest that effective AI governance requires a shift from restrictive, compliance-focused approaches towards enabling and adaptive strategies. Public sector organisations should provide approved AI tools, clear usage guidelines and controlled environments (e.g. sandboxes) to support transparent and responsible use. Governance mechanisms should be embedded within everyday workflows and supported by strong human oversight, review practices and accountability structures. Importantly, organisations must recognise the increased effort required for verification and invest in training and organisational readiness. Treating governance as an ongoing capability rather than a static policy is essential for managing AI responsibly. The findings highlight important implications for accountability, transparency and public trust in AI-enabled government. By showing that AI use intensifies rather than replaces human judgment, the study underscores the continued centrality of human responsibility in public decision-making. However, the persistence of informal and “shadow” AI practices raises concerns about uneven access, reduced transparency and potential risks to fairness and oversight. The results suggest that enabling and practice-informed governance approaches are essential to ensure that AI adoption in the public sector supports legitimacy, inclusiveness and responsible innovation in the service of broader societal values. This study contributes to the information systems literature by conceptualising AI governance as a process of dynamic alignment between organisational practices and governance mechanisms. It extends existing work by providing an empirically grounded account of how governance co-evolves with AI use under conditions of institutional constraint and limited organisational readiness. The findings highlight the importance of socio-technical dynamics, human-in-the-loop oversight and adaptive governance approaches in achieving responsible AI in the public sector.