Year: 2025
Type: article
Abstract: Artificial Intelligence is increasingly central to modern cybersecurity, offering unprecedented capabilities in detecting, analysing and responding to threats at machine speed and scale.This paper presents a structured review of the state of AI-driven cybersecurity, focusing on how supervised learning, unsupervised anomaly detection, deep learning and reinforcement learning are being operationalised across threat detection, phishing prevention, user behaviour analytics and autonomous response systems.Case studies, including Google's phishing detection and Microsoft's Security Copilot, illustrate AI's role in enhancing both efficiency and accuracy in cyber defence.However, the integration of AI introduces new risks such as adversarial attacks, model evasion, false positives, explainability gaps and data scarcity.We explore these challenges alongside the emerging AI-versus-AI threat landscape, where malicious actors also weaponize AI to evade detection and automate attacks.In parallel, we assess evolving policy and governance frameworks such as the EU AI Act and NIST's AI Risk Management Framework, highlighting the importance of transparency, accountability and privacy in deploying AI responsibly.Finally, we outline future directions including the rise of predictive cybersecurity, AI and blockchain convergence for distributed trust and the need for adversarial resilience in model design.We argue that securing the digital future requires not only technical innovation but also cross-sector collaboration to ensure AI systems are robust, interpretable and ethically aligned.
Institutions: University of Salford, University of Plymouth
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4412190203