Governing Artificial Intelligence in Nursing Practice in Low- and Middle-Income Countries: A Critical Integrative Review and Nurse-Centred Sociotechnical Framework

Background: Artificial intelligence (AI) is increasingly promoted as a means of strengthening clinical decision-making, documentation, triage, surveillance, remote monitoring, and health-system efficiency. In low- and middle-income countries (LMICs), where nurses often sustain frontline care amid workforce shortages, fragile infrastructure, limited specialist access, and constrained public resources, AI could support safer and more timely care. Yet its implementation also risks reproducing inequity when tools are imported, poorly validated, culturally mismatched, weakly regulated, or introduced without meaningful nursing participation. Aim: To synthesize current evidence and policy guidance on barriers and enabling conditions for AI implementation in LMIC nursing practice and to develop a nurse-centred sociotechnical governance framework for safe, equitable, and context-responsive adoption. Design: Critical integrative review with framework synthesis. Methods: Peer-reviewed literature, AI reporting standards, implementation science frameworks, nursing workforce evidence, and authoritative global policy documents were synthesized. Sociotechnical systems theory provided the primary analytic lens, supported by value-sensitive design, technology acceptance constructs, implementation outcomes, and prior nurse-centred AI readiness scholarship. Evidence was charted across infrastructure, workforce, data, governance, cultural-linguistic, and workflow domains and interpreted at micro, meso, and macro levels. Results: Six interdependent implementation domains were identified: digital infrastructure and health information systems; nursing AI literacy and professional agency; data quality, representativeness, and sovereignty; ethical, legal, and regulatory governance; cultural-linguistic localization; and human-AI collaboration within nursing workflows. The synthesis shows that AI risk in LMIC nursing is not reducible to technical error. It emerges from interactions among infrastructure instability, constrained labour conditions, imported or poorly validated models, data asymmetries, weak procurement safeguards, regulatory gaps, and the marginalization of nurses from design and oversight. Conclusion: AI implementation in LMIC nursing should be governed as sociotechnical health-system transformation, not technology transfer. Safe adoption requires local validation, nurse participation, data justice, accountable procurement, proportional regulation, cultural-linguistic adaptation, and post-deployment learning governance. Impact: The proposed framework offers policymakers, nursing leaders, educators, regulators, developers, funders, and global health partners a practical structure for integrating AI without weakening nursing judgement, patient dignity, or health equity.