Artificial Intelligence (AI) security research is promising and highly valuable in the current decade. In particular, deep neural network (DNN) security is receiving increased attention. Although DNNs have recently emerged as a prominent tool for addressing complex challenges across various machine learning (ML) tasks and DNNs stand out as the most widely employed, as well as holding a significant share in both research and industry, DNNs exhibit vulnerabilities to adversarial attacks where slight but intentional perturbations can deceive DNNs models. Consequently, several studies have proposed that DNNs are exposed to new attacks. Given the increasing prevalence of these attacks, researchers need to explore countermeasures that mitigate the associated risks and enhance the reliability of adapting DNNs to various critical applications. As a result, DNNs have been protected against adversarial attacks using a variety of defense mechanisms. Our primary focus is DNN as a foundational technology across all ML tasks. In this work, we comprehensively survey and present the latest research on DNN security based on various ML tasks, highlighting the adversarial attacks that cause DNNs to fail and the defense strategies that protect the DNNs. We review, explore, and elucidate the operational mechanisms of prevailing adversarial attacks and defense mechanisms applicable to all ML tasks utilizing DNN. Our review presents a detailed taxonomy for attacker and defender problems, providing a comprehensive and robust review of most state-of-the-art attacks and defenses in recent years. Additionally, we thoroughly examine the most recent systematic review concerning the measures used to evaluate the success of attack or defense methods. Finally, we address current challenges and open issues in this field and future research directions.
28 February 2026 in International Journal of Advances in Intelligent Informatics
Clean-label poisoning attacks pose a stealthy and potent threat to deep neural networks (DNNs), particularly when models rely on publicly available or outsourced training data. Among these attacks, the Bullseye Polytope method is highly transferable and can evade state-of-the-art defenses such as deep k-NN. To counter this, we propose Poison Image Traceback via Feature Clustering (PIFC-CLD), a novel forensic approach that leverages Euclidean norm distances to detect and trace clean-label attacks in DNNs. PIFC exploits the geometric consistency of feature representations to identify poisoned samples responsible for model misclassifications. Unlike traditional majority-vote-based defenses, PIFC-CLD performs clustering in feature space and detects poisoned samples based on their proximity to misclassified targets using Euclidean distance. We evaluate our approach under Bullseye Polytope attack scenarios using the CIFAR-10 dataset and WideResNet architectures. PIFC-CLD achieves 99% precision, 95% recall, and a 96% F1 score at k = 25 and ε = 0.2, demonstrating robust performance against Bullseye Polytope attacks. Furthermore, our algorithm exhibits strong resilience to parameter variations while minimizing false positives and preserving model integrity. This work bridges the gap between digital forensics and adversarial machine learning, offering a lightweight, model-agnostic, and interpretable solution for secure model training in adversarial environments.