ZERO TRUST IN PRACTICE: AI-AUGMENTED ACCESS CONTROL FOR CRITICAL GOVERNMENT NETWORKS

Abstract

The increased complexity of the current government networks requires dynamic and open-minded cybersecurity systems capable of interoperating with distributed and legacy systems. Conventional perimeter-based security frameworks are no longer effective to deal with advanced persistent threats or insider abuse, especially in multi-agency settings where data confidentiality and interoperability are paramount. The proposed research suggests an AI-enhanced Zero Trust Architecture (ZTA) incorporating user-behaviour analytics, federated learning (FL), and explainable artificial intelligence (XAI) to provide dynamic and auditable access to important government systems. The architecture assesses contextual risk on a continuous basis on the basis of behavioural deviations and makes adaptive access decisions using an intelligent policy engine. The model was trained and validated on five simulation nodes of the agency to simulate distributed governmental environments using the User Activity Anomaly Detection Dataset. Federated learning ensured data sovereignty by summing model parameters as opposed to raw data, whereas SHAP-based explainability offered interpretable information about each access decision. Experimental findings showed that the overall accuracy was 92%, 0.89 F1-score and 0.94 ROC-AUC, with only slight performance loss as compared to a centralised model, yet full privacy protection. A 32% decrease in undetected insider incidents and enhanced compliance auditing was achieved with the system due to transparent decision logic. Such results highlight the possibility of integrating smart, understandable and cooperative security controls into the running government architectures. Besides, the strategy is consistent with the overall Computer Integrated Manufacturing Systems (CIMS) vision in terms of the demonstration of how intelligent integration, automation and human control can conglomerate into the computer-related governance systems that are resilient.