Zero trust security architecture enhanced with federated learning for modern network environments

Abstract

The fast-paced digitization and the adoption of cloud-based services have greatly widened the enterprise attack surface and put the networks at risk from internal vulnerabilities as well as external attacks. The traditional model of perimeter-based security cannot adequately protect the modern distributed infrastructure where hybrid work, Bring-Your-Own-Device (BYOD) policies, and Internet-of-Things (IoT) ecosystems are prevalent. This paper presents an approach to a Zero Trust Security Architecture (ZTA) by integrating Federated Learning (FL) in order to provide adaptive, privacy-preserving, and intelligent network protection. The proposed solution allows the ZTA’s “never trust, always verify” principle to be enforced while implementing federated learning in order to train AI models collaboratively amongst multiple distributed nodes and without exchanging raw data, thus ensuring confidentiality. Centralized federated learning is used for secure authentication, and Distributed federated learning systems are used for independent constant diagnostics and policy updates. The proposed mathematical formulation and performance comparisons have shown that the combined frameworks can enhance data privacy, scalability, and resiliency while aligning with contemporary Cyber Security Regulatory compliance. We also provide the groundwork for integrating with blockchain and edge optimized AI systems in future work.