AI-Driven Personalized Decision Support for Secure and Adaptive Digital Services

Abstract

Abstract Neurosymbolic AI and Zero Trust Architecture (ZTA) are increasingly proposed together as a combined approach to securing cloud-edge IoT continuums in federal and critical infrastructure sectors, on the premise that neurosymbolic systems can supply the explainable, robust threat detection that continuous-verification security models require (Akter, Chowdhury, Munny, Haque, Mambetaliev, and Karshiboev, 2026). This paper evaluates that premise through a critical review of the peer-reviewed literature on Zero Trust Architecture maturity and neurosymbolic AI for network security. Drawing on a 74-study systematic review of ZTA implementation across domains (Mushtaq, Mohsin, and Mushtaq, 2025), this paper finds that ZTA research attention is substantially concentrated in cloud computing (24 of 74 reviewed studies) relative to the Internet of Things specifically (11 studies), and that no reviewed ZTA implementation satisfies all core cybersecurity dimensions, authentication, authorization, access control, encryption, auditing, and environmental perception in a holistic manner, with auditing, orchestration, and environmental perception consistently the least mature components across domains. Separately, the neurosymbolic AI security literature shows genuine and growing technical capability for combining deep learning's pattern recognition with symbolic reasoning's interpretability in network intrusion detection, but this capability has been demonstrated primarily at the detection layer, not yet at the orchestration and policy-enforcement layer that a full zero trust deployment requires. This paper proposes a deployment-readiness framework mapping ZTA maturity against neurosymbolic explainability adequacy, and argues that claims of neurosymbolic-AI-enabled zero trust orchestration for IoT specifically should be read against a research base that is real and growing but still concentrated in adjacent domains rather than IoT-native, multi-tenant, resource-constrained deployment. This paper does not present original system implementation results; its contribution is a synthesis grounded in independently verified sources.