Explainable Artificial Intelligence at the Intersection of Security, Finance, and Governance: A Unified Review with Novel Conceptual Extensions
DOI:
https://doi.org/10.54692/lgurjcsit.v10i1.204Keywords:
Explainable Artificial Intelligence, Financial AI, SHAPAbstract
Explainable Artificial Intelligence (XAI) has transitioned from a technical curiosity to a fundamental aspect of trustworthy, accountable, and compliant use of artificial intelligence. While XAI literature is increasing, current reviews approach XAI either as a technical issue or as a domain-specific issue, but do not offer an integrated, multi-sectoral analysis that spans security, finance, and governance. This paper offers an original, integrated review of two parallel threads in XAI research: the wide range of XAI methods and their focused deployment in finance. We propose three novel conceptual frameworks: (i) a Layered Explainability Gradient (LEG) model building on existing concepts of marginal transparency and marginal interpretability to develop a multi-dimensional decision support model; (ii) a Cross-Domain XAI Applicability Matrix (CXAM) that aligns XAI methods with domain-specific risk profiles and regulatory frameworks; and (iii) a Governance-Integrated Explainability Roadmap (GIER) extending the 2025-2030 research roadmap with explicit consideration of the EU AI Act, GDPR, and future international norms. Our review of 138 finance-centred studies, complemented by a thematic review of general XAI developments, reveals persistent deficiencies in the standardization of evaluation metrics, user-adapted explanation granularity, privacy-preserving explainability and adversarial robustness. Our analysis shows that SHAP and LIME are still the most prevalent but not the most effective explanatory methods, and that multi-layered, user-adaptive and privacy-preserving explainability models are needed. This review provides a clear, practical guide for researchers, practitioners and policy makers to drive explainable and trustworthy AI.