The increasing use of Machine Learning (ML) in critical domains like that of health, finance, and cybersecurity has made it increasingly challenging to ensure security and robustness of its models. The area of focus includes adversarial attacks, where carefully crafted perturbations of input data lead to incorrect or misleading predictions. In this paper, we explore the role of explainable AI (XAI) in securing and assuring robustness of ML systems against such attacks. We look at identifying vulnerabilities, adversarial patterns for detection, and increased robustness. The models used are Local interpretable model-agnostic explanations (LIME), SHapley Additive exPlanations (SHAP), and attention-based models and illustrate how XAI can deal with security vulnerabilities in protecting ML models against adversarial attacks. Additionally, we discuss security-related problems faced by ML models such as that of adversarial perturbations, data poisoning, and model inversion attacks; insight into how XAI can dispel such risks by providing interpretability which in turn could empower remedying of such adversarial patterns in the future. The paper elucidates how XAI can bolster trustworthy AI systems by empowering them to secure more effective security strategies which comply with ethical AI. The paper concludes with future directions to render XAI robust against adversarial attacks, to enhance it with real-time operability by optimizing its computational efficiency, and to integrate it with new strategies for learning deep neural networks.