Artificial Intelligence for Cyber Threat Detection in the Banking Sector: A Systematic Review

Abstract

The rapid digital transformation of the banking sector has significantly increased its exposure to sophisticated cyber threats, necessitating advanced detection mechanisms beyond traditional rule-based systems. This systematic review examines the application of Artificial Intelligence (AI) for cyber threat detection in the banking sector, synthesizing evidence from 41 peer-reviewed studies published between 2014 and 2025. The review follows PRISMA guidelines for study identification, screening, and inclusion. Findings reveal that Machine Learning (ML) and Deep Learning (DL) techniques, including hybrid models, deep neural networks, and autoencoders, have demonstrated detection accuracies up to 99.98% for network intrusions and significant reductions in false positive rates. Key applications include fraud detection, real-time transaction monitoring, behavioral biometrics-based authentication, and anomaly detection in financial networks. However, several barriers impede widespread adoption: system complexity, lack of AI-proficient personnel, regulatory compliance challenges, model interpretability concerns, and integration difficulties with legacy banking infrastructure. The review identifies that socio-technical challenges, including shadow AI usage and limited audit readiness, represent critical gaps in current research. This review contributes a comprehensive taxonomy of AI techniques for banking cybersecurity and provides actionable recommendations for financial institutions seeking to implement AI-driven threat detection systems. Future research directions include privacy-preserving techniques such as federated learning, adversarial robustness, and explainable AI frameworks tailored to regulatory requirements.