An <scp>AI</scp> ‐Based Framework for Secure and Transparent Banking: Integrating Adversarial Robustness, Interpretability, and Organizational Modeling

ABSTRACT This paper presents an integrated AI‐based cybersecurity framework for banking institutions that addresses technical and organizational challenges. Unlike traditional security systems that rely solely on detection accuracy, the proposed framework unifies adversarial robustness, explainable AI (XAI), and agent‐based modeling (ABM) to detect and simulate responses to cyber threats. It emphasizes the role of knowledge management by embedding interpretability tools, such as SHAP and LIME, and modeling human decision‐making under threat conditions. We validated the framework using adversarial simulations with FGSM and PGD, SHAP‐based interpretability testing, and ABM‐driven response modeling. The approach demonstrates favorable results in improving transparency, resilience, and adaptive response capacities. Our model achieved 94.1% accuracy and 92.8% F1 Score while maintaining 89% SHAP feature stability against FGSM and PGD attacks.