AGENTIC AI FOR AUTONOMOUS CYBER-DEFENSE: A HUMAN-IN-THE-LOOP FRAMEWORK FOR PAKISTAN’S CRITICAL DIGITAL INFRASTRUCTURE

Abstract

The increasing sophistication, speed, and adaptability of cyber threats have created a critical need for intelligent and autonomous defense mechanisms for critical digital infrastructure. This study developed a Human-in-the-Loop (HITL) framework for Agentic Artificial Intelligence (AI)-enabled autonomous cyber defense within Pakistan’s critical digital infrastructure. A quantitative cross-sectional research design was employed, and data were collected from 267 cybersecurity and information-technology professionals through a structured questionnaire. The findings indicated that Agentic AI Capability significantly predicted autonomous cyber-defense effectiveness (β = .32, p < .001), while Human-in-the-Loop Oversight (β = .27, p < .001), Cybersecurity Risk Management (β = .29, p < .001), and Explainability and Accountability (β = .18, p = .001) also demonstrated significant positive effects. Autonomous cyber-defense effectiveness was further strongly associated with cyber resilience (β = .52, p < .001). The results supported a risk-adaptive autonomy approach in which low-risk and reversible cybersecurity actions were automated, whereas high-impact or uncertain decisions remained subject to human authorization. The study concluded that integrating agentic AI with structured human oversight, explainability, risk management, and accountability could enhance the security and resilience of Pakistan’s critical digital infrastructure. The proposed framework provides a foundation for responsible deployment and future empirical validation of autonomous cyber-defense systems.