Banking sector is crucial for managing financial transactions and protecting customer assets, but it faces serious security threats from hackers and fraudsters. Cyber-attack poses a significant risk by decrypting bank’s data and demanding payment, which can disrupt operations and steal sensitive information. Conventional cyber threat detection techniques such as rule and signature-based systems, frequently fail to detect threats and may not be able to sufficiently address evolving and dynamic attacks. To address these concerns, this paper proposes an optimized Bi-GRU-based cyber-attack detection and N-th degree Truncated polynomial Ring Units (NTRU) cryptography to enhance the security of banking information. Data relevant to packet transmission is initially collected from Standard source and pre-processed using a bi-objective nearest neighbour’s imputation method to accurately fill missing values, along with symbolic feature normalization for uniform scaling. After pre-processing, features are extracted using a stacked sparse autoencoder, and XGBoost model is applied for feature selection to improve stability and reduce data complexity. The selected features are then fed into an optimized Bi-GRU classifier (O-Bi-GRU) to classify it as attack or non-attack. NTRU cryptography technique is used in banking networks to provide double layer protection, which improves the security of non-attack data. The system demonstrates superior computational performance compared to the current methods based on performance evaluations and security analyses. Based on the experimental study, the suggested method achieves 98% accuracy, 96% precision, 3% false positive rate (FPR), 97% specificity, 96% negative predictive value (NPV), and encryption times of 4.86 ms, respectively. Consequently, the O-BiGRU-based cyber threat detection combined with NTRU cryptography enhances the security of the banking sector through mitigating threats effectively.