Between Protection and Control: A Scoping Review of Artificial Intelligence Applications and Ethical Governance In Refugee and Immigrant Services

Objective: AI systems are increasingly embedded in refugee and immigrant services that embody an inherent tension between protection and control – promising access and support while simultaneously exposing individuals to surveillance, gatekeeping, and reduced contestability. As AI mediates services across migration contexts, there is a growing need to examine not only where these technologies are deployed but also how ethical considerations are addressed in practice. This scoping review maps AI applications in refugee and immigrant identification, screening, and support services across migration stages, and examines how ethical governance is addressed in existing research.

Method: We searched four databases and screened 4,026 records following PRISMA-ScR guidelines. Fourteen studies met the inclusion criteria: AI systems embedded directly into services serving refugee and immigrant populations. We applied a conceptual framework adapted from Richardson et al.'s (2022) Framework for Digital Health Equity, organizing AI applications across two migration stages – During Migration and Post-Migration – and four socioecological levels, with an Ethical Governance row drawing on Matlin et al. (2025) and the UNHCR AI Approach (2025).

Results: Studies were organized into two migration stages: During Migration (n = 4), encompassing pre-entry health screening and admission control as gatekeeping, and detection; and Post-Migration (n = 10), encompassing resettlement location assignment, surveillance, health and psychosocial well-being, and service navigation, integration, and employment support. Mapping studies onto the conceptual framework revealed marked differences in ethical governance engagement across migration stages and socioecological levels. No During Migration study explicitly addressed any ethical governance component; ethical discussion was confined to procedural safeguards such as IRB approval and regulatory compliance. Among Post-Migration studies, substantive engagement was concentrated in five studies, with components spanning individual-level considerations (informed consent, autonomy, right to contest decisions), interpersonal dynamics (transparency with clients, human oversight of AI outputs), and community-level processes (community involvement in AI design, feedback mechanisms). Human rights-based AI governance was not explicitly addressed by any included study. Across both stages, affected populations had minimal involvement in system design or governance.

Conclusions: AI ethics in refugee and immigrant services is unevenly addressed, often reduced to procedural safeguards, leaving refugees and immigrants with limited agency. Future research should move beyond procedural ethics toward approaches that recognize refugees and immigrants not only as populations to be protected, but as central to ethical governance, including in the design and oversight of AI-enabled services.