This study presents an innovative AI-driven approach designed to enhance security measures against threats in IoT banking systems, addressing the increasing vulnerabilities in this critical domain. The newly developed method achieves a precision of 0.88 and a balanced recall of 0.79, effectively strengthening defenses against cyberthreats. Utilizing a sophisticated deep neural network architecture called pointer networks, this mechanism adjusts dynamically, ensuring high accuracy in identifying threats (precision) while comprehensively covering potential risks (recall), resulting in an overall F1 score of 0.83. Rigorous evaluations tailored to specific threat scenarios demonstrate its versatility, exhibiting superior performance in detecting malware (precision: 0.89, recall: 0.82, F1 score: 0.85), countering denial-of-service (DoS) attacks (precision: 0.87, recall: 0.78, F1 score: 0.82), and thwarting unauthorized access attempts (precision: 0.90, recall: 0.81, F1 score: 0.85). In various threat assessments, the model performs effectively, showcasing particular strength in identifying malware (precision: 0.89, recall: 0.82, F1 score: 0.85), mitigating denial-of-service (DoS) incidents (precision: 0.87, recall: 0.78, F1 score: 0.82), and preventing unauthorized access (precision: 0.90, recall: 0.81, F1 score: 0.85).