Empowering Cybersecurity: Evaluating Agentic AI Frameworks for Threats Detection and Adaptive Incident Response

Abstract

This paper presents a comprehensive review of Agentic AI frameworks in cybersecurity, with particular emphasis on autonomous threat detection and adaptive incident response. As cyber threats continue to evolve in complexity and scale, traditional rule-based security mechanisms are becoming increasingly ineffective in responding to dynamic and sophisticated attacks. Agentic AI introduces a transformative approach by integrating real-time monitoring, continuous learning, autonomous reasoning, and adaptive decision-making into cybersecurity operations. The review examines recent advances in areas such as autonomous threat detection, SOC automation, adaptive response systems, governance, explainability, and adversarial risks. The findings indicate that Agentic AI significantly improves detection accuracy, accelerates incident response, reduces false positives, and enhances overall cyber resilience through technologies including machine learning, deep learning, reinforcement learning, and multi-agent systems. However, despite these promising developments, current research remains fragmented and faces several limitations, including limited real-world validation, insufficient explainability, governance challenges, and vulnerabilities such as adversarial attacks, data poisoning, and model manipulation. The study highlights the need for secure, transparent, and human-centered deployment strategies, as well as standardized evaluation frameworks and stronger human-AI collaboration models. Overall, Agentic AI represents a promising paradigm for building more intelligent, adaptive, and resilient cybersecurity systems capable of addressing the challenges of modern digital environments.