Explainable artificial intelligence for secure and privacy-preserving 5 G networks

Explainable artificial intelligence (XAI) is increasingly relevant to the security, privacy, and governance of AI-enabled 5 G and emerging 6 G networks. However, existing studies remain fragmented across wireless tasks, data modalities, network planes, and evaluation criteria, making it difficult to determine when an explanation is operationally useful, sufficiently faithful, and feasible under latency and privacy constraints. This scoping review maps and synthesizes the literature on XAI for wireless security, resource management, Open RAN, network slicing, edge intelligence, and Zero Trust architectures. It introduces marginal transparency and marginal interpretability as a trade-off-aware conceptual framing for reasoning about diminishing explanatory returns as complexity, latency, cognitive burden, and privacy exposure increase. The review further develops a multidimensional taxonomy based on explanation timing, scope, model dependence, computational profile, deployment plane, and risk characteristics. Particular attention is given to explanation faithfulness, stability, adversarial manipulation, privacy leakage, and the emerging use of large language models as explanation generators and network decision-support agents. Based on the reviewed evidence, we identify methodological and deployment gaps and present a research agenda for developing reproducible, latency-aware, privacy-preserving, and operationally grounded XAI for next-generation wireless networks.