From innovation to implementation: a global scoping review of artificial intelligence applications, governance, and challenges in humanitarian contexts.

Background: Artificial intelligence (AI) is increasingly being explored as a tool for supporting humanitarian action through hazards and crisis prediction, real time monitoring of evolving crisis conditions and humanitarian needs, resource allocation, and decision-making. However, evidence on how AI is implemented, evaluated, and governed across humanitarian contexts remains fragmented. In this scoping review, we mapped the global evidence on the application of AI in humanitarian settings, including its distribution across the emergency management cycle, deployment maturity, and evaluation. We also identified any governance considerations, as well as enabling factors for and challenges to its implementation, with particular attention to the Middle East and North Africa (MENA) region, Gulf Cooperation Council countries (GCC), and Qatar. Methods: The methodology of this scoping review followed the framework set by Arksey and O'Malley, which comprises the formulation of the research question, selection of records, data charting, and synthesis of findings. We searched six databases (ProQuest, IEEE Xplore, ScienceDirect, Embase, Web of Science, and Scopus), queried Google Scholar, and ran targeted grey literature searches through the webpages of major humanitarian organisations for records reporting on the application, implementation, or governance of AI in humanitarian contexts. We descriptively and narratively synthesised the extracted data. Results: We included 57 scholarly publications and 17 grey literature reports. Application of AI was most frequently reported in sudden-onset natural disasters, particularly floods and earthquakes, and was commonly used to support disaster monitoring, crisis informatics, search-and-rescue, logistics optimisation, hazard prediction, and geospatial analysis. Deep learning and machine learning were the most common model families. Across the emergency management cycle, most applications of AI focused on preparedness and response, while recovery was rarely addressed. Geographically, most evidence originated from Asia, while research from the MENA and GCC regions remained limited. Most reported systems remained prototypes or retrospectively validated models, with relatively few seeing operational deployment or field implementation. Governance issues were discussed more extensively in grey literature than in scholarly publications. Community participation, equity considerations, and formal governance frameworks were inconsistently reported. Conclusions: AI is increasingly being applied in humanitarian contexts, but it is heterogeneously distributed across regions, humanitarian contexts, and emergency management. Greater emphasis is needed on real-world deployment, implementation-focused evaluation, participatory approaches, and governance frameworks that support responsible, equitable, and context-sensitive AI adoption.