From Detection to Resilience: A Review of Explainable AI Trust and Real-Time Response in AI-Driven Bank Fraud Detection
Year: 2026
Type: article
Abstract: Background: Banks use machine learning to flag fraudulent transactions at a scale that rule-based systems cannot match. Detection accuracy has received most of the research attention. Much less is known about whether better detection produces a more resilient institution.Purpose: This review examines how AI-driven fraud detection capability relates to the cybersecurity resilience of banks, and what role explainability-based trust and real-time response capability play in that relationship.Method: A narrative review of research on statistical and machine-learning fraud detection, explainable AI, human trust in automation, cyber resilience, and partial least squares structural equation modeling (PLS-SEM) studies in banking. A recent U.S. banking survey study (Rimi et al., 2025) is used as an applied reference point.Findings: Three themes emerge. First, detection capability is well documented, but its link to organizational resilience is mostly assumed rather than tested. Second, explainability affects outcomes through human trust and reliance, not through model performance alone. Third, response speed is a weak predictor of resilience when it is not tied to recovery, governance, and learning. The reference survey study found that detection capability predicted resilience directly (β = 0.444) and through explainable AI trust (indirect β = 0.206), while real-time response capability had no significant effect (β = −0.002, p = 0.959).Conclusion: Banks should treat explainability, analyst training, and recovery integration as part of the fraud detection investment. Future work needs longitudinal designs, objective resilience measures, and tests of governance as a moderator.
Source: Figshare
Author: Shafiqur Rahman
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7234128437
