| Abstract: |
Cyber threats are becoming more sophisticated, and digital infrastructures are growing more complex. This has increased the use of Artificial Intelligence (AI) agents for autonomous cybersecurity operations. Research so far has examined several ways to deploy AI agents in cybersecurity. These include centralized architectures, multi-agent systems, Retrieval-Augmented Generation-based frameworks, architectures using the Model Context Protocol, and hybrid models. Hybrid models can combine reasoning, retrieval, orchestration, and tool integration capabilities. These architectures have demonstrated significant potential to enhance threat detection, incident response, and threat intelligence analysis, as well as to automate Security Operations Center tasks. However, deploying these approaches faces several challenges. These include hallucinations in Large Language Models, prompt injection attacks, adversarial manipulation, scalability limitations, interoperability challenges, governance issues, and a lack of standardized datasets and benchmarks. This review paper analyzes AI agents in cybersecurity. It covers the main architectural types, their applications, security challenges, available cybersecurity datasets, and development frameworks. The study also compares current approaches and highlights key research directions for building secure, scalable, and trustworthy autonomous cyber defense systems. |