Authors: Rohit Agnihotri
Abstract: The exponential growth of digital enterprise environments has rendered traditional, rule-based access governance frameworks operationally inadequate. Static policies, periodic access reviews, and reactive threat response mechanisms cannot match the velocity, complexity, or adaptability of modern identity ecosystems or the adversaries that target them. This article examines the emerging paradigm of autonomous access governance through self-learning Artificial Intelligence (AI) agents that provide real-time policy enforcement and continuous behavioral modeling. Drawing on empirical performance data, agent-based system architectures, and regulatory compliance frameworks, we demonstrate that self-learning AI agents achieve policy decision accuracy of up to 96.8% after 30 learning cycles (a 69% improvement over static rule-based systems) while reducing policy enforcement latency from thousands of seconds to sub-second response times. We further document a 94.4% reduction in monthly access policy violations over twelve months of post-deployment operation. We introduce a five-component Autonomous Access Governance Architecture (AAGA) and a four-tier Governance Maturity Model to guide practical implementation. The article concludes by identifying critical open research challenges in agent auditability, adversarial robustness, and the governance of recursive self-modification in deployed AI governance agents.
DOI: https://doi.org/10.5281/zenodo.23561577
