AI-Powered Cybersecurity: Emerging Trends and Challenges – A Narrative Review

Abstract

This study is driven by the fact that artificial intelligence (AI) technologies are quickly changing the landscape of intrusion detection, malware analysis and mitigation of advanced cyber-attacks making cybersecurity a domain most impacted by this revolution. In order to provide an objective overview on AI uses in cybersecurity, a systematic literature review is conducted. In this review, such algorithms like machine learning or deep learning have been compared to hybrid architectures (CNN–LSTM or graphs neural networks) regarding accuracy, interpretability, efficiency and generalizability towards real-life scenarios. It digs deeper into recent research on benchmark datasets and evaluation protocols to reveal the gap between academic experiments, versus actual deployment. The hybrid, task-specific and cross-task feature-learning models always outperformed standalone methods but are computational expensive and lack interpretability. Benchmark datasets also led to the divergence between laboratory results and real-world cybersecurity scenarios due to the over-dependence on benchmarks. The core challenges are imbalance data, partial explainability, adversarial attacks and poor transferability across domains. Future research should study lightweight, interpretable hybrid AI systems with unified learning frameworks and graph-based models dedicated to edge & resource-limited environments but takes into account the trade-off between interpretability and efficiency striding towards real-world applications deployment contexts under constraints. The synthesis of empirical evidence, along with gaps in research, offers guidelines for effective design and deployment of scalable, performant cybersecurity systems aimed towards real-world deployments that support scalability against different environments to improve the detection capabilities at scale.