Adversarial attacks pose a significant challenge to the robustness of deep neural networks (DNNs), particularly in sensitive applications such as image classification. We propose a comprehensive approach to enhance the resilience of DNNs by integrating three advanced techniques: Robust Artificial Immune Learning System (RAILS), Jacobian regularization, and label smoothing. This combination of methods specifically addresses gaps in existing defenses by introducing an innovative multi-step process through RAILS, which works synergistically with Jacobian regularization and label smoothing to provide a balanced improvement in both adversarial robustness and model accuracy. Label smoothing reduces overconfidence in predictions, ensuring improved generalization, while Jacobian regularization reduces model sensitivity to input variations, thereby increasing robustness. RAILS, inspired by the immune system, introduces a multi-step process involving sensing, flocking, affinity maturation, and consensus, fostering adversarially robust learning in a way that existing approaches do not. Experimental results on the MNIST dataset demonstrate a 46.13% improvement in testing accuracy, from 52.80 to 98.93%, by increasing the number of hidden layers, with a decrease in adversarial accuracy from 96.66 to 88.59%. This trade-off between testing accuracy and adversarial robustness was effectively managed by incorporating regularization and augmentation techniques, ensuring that the model remains both highly accurate on clean data and resilient to adversarial attacks. Finally, we outline potential avenues for further research to extend these techniques to more complex datasets and real-world scenarios, particularly emphasizing the novelty of combining RAILS with established methods for adversarial defense, which provides a robust foundation for future advancements in DNN robustness.