Risk-Aware Agentic AI for Zero-Trust SD-WAN in SaaS Enterprises

Abstract

The rapid adoption of Software-as-a-Service (SaaS) applications has reshaped enterprise IT ecosystems, simultaneously expanding the attack surface and challenging traditional security models. While Software-Defined Wide Area Networking (SD-WAN) offers flexibility and scalability, its integration with Zero-Trust principles often encounters limitations in adaptability and administrative overhead. This study proposes a novel framework that embeds risk-aware agentic artificial intelligence (AI) into Zero-Trust SD-WAN to deliver dynamic, autonomous, and context-sensitive security. Using reinforcement learning as the core mechanism, the framework autonomously evaluates user and device risks, adapts access policies in real time, and responds proactively to emerging threats. Experimental simulations under varying traffic loads and threat scenarios reveal that the approach achieves detection rates exceeding 90%, reduces false positives, and maintains network performance with packet delivery ratios above 95%. Moreover, the system balances stringent access enforcement with user experience, ensuring high satisfaction among trusted users while containing insider misuse. Reinforcement learning models demonstrated superior efficiency, achieving faster convergence, lower resource utilization, and higher predictive accuracy compared to supervised and unsupervised alternatives. These findings underscore the potential of risk-aware agentic AI as a scalable and future-ready enabler for intelligent Zero-Trust SD-WAN deployments in SaaS enterprises.