Artificial Intelligence in Auditing and Banking Risk Management: A Bibliometric Mapping of Explainability, Fraud Detection, and Regulatory Compliance

Résumé

Artificial intelligence is already present in a wide range of banking and audit activities, but the literature around these uses is still split across several research conversations. Banking studies usually concentrate on prediction, fraud monitoring, cybersecurity, and risk management. Audit research asks a different question: can the outputs be documented, challenged, controlled, and used in professional judgment? Work on explainable AI adds another layer, namely whether complex model outputs can be understood well enough to support scrutiny. We focus on the overlap among these streams rather than on AI in banking as a whole. To date, this overlap has not been the subject of a dedicated bibliometric mapping, which is the gap this study addresses. Four questions guide the analysis: how have scientific production and influence evolved, which concepts structure the field, which references does it draw on, and how do authors and countries collaborate? The dataset contains 43 Scopus-indexed documents published between 2014 and 2026. Using Bibliometrix/Biblioshiny in R, we combined performance analysis, thematic mapping and evolution, multiple correspondence analysis (MCA), co-citation, historiography, and collaboration networks. Screening was intentionally selective. Prediction-only papers were excluded when assurance, control, auditability, or related oversight issues were not substantive. Thirty-four papers (79.1%) were published between 2024 and 2026. AI and machine learning form the technical core of the literature, with explainability, financial risk management, fraud detection, cybersecurity, and regulatory compliance recurring around it. MCA identifies three clusters: a broad AI-risk-fraud-explainability cluster, a smaller audit-compliance-fintech cluster, and a specialized blockchain-regulatory-compliance cluster. Citation and collaboration patterns are less settled: co-citation communities remain fragmented and direct citation chains are short. The literature therefore looks less like a finished theoretical framework than a set of technical and governance concerns that are beginning to converge. We interpret this convergence through the notion of AI-enabled financial risk governance and assurance, which distinguishes AI capability from the assurance capability needed to audit and control its use. In that setting, explainability and auditability matter because even a high-performing model still has to withstand professional challenge and regulatory scrutiny.

JEL Classification: G21; G32; M42; O33

Paper type:Empirical Research (bibliometric study)