An adaptive behavioural zero-trust security framework with proactive cryptographic caching for continuous authentication in 5G non-terrestrial networks

Abstract 5G non-terrestrial networks (NTNs) are key to future global broadband and military communications, but their high latency and dynamic topologies require security approaches beyond perimeter defences and static authentication. To overcome these vulnerabilities, we propose an adaptive behavioural zero trust security framework (BZTSF) for continuous 5G NTN authentication. Our method continuously evaluates node trustworthiness using network signal information processed by an isolation forest machine learning algorithm. A proactive cryptographic caching mechanism reduces re-authentication latency to 7.2 ms with 99.5% session continuity (evaluated under warm-cache rule-based conditions), compared to traditional 1600 ms handover peaks. When subsequently integrated with the ML engine, the framework demonstrated a highly efficient signalling overhead of 0.75%, well below the 8% target. A policy decision point engine instantly translates anomaly scores into access control decisions. Evaluating 50,000 simulated NTN telemetry samples (4:1 class imbalance) yielded 88.00% accuracy, 0.9856 average precision, and a 0.9969 ROC AUC (area under the receiver operating characteristic curve). Within the simulated environment, the framework delivered 100% live spoofing detection and 1.0000 attack recall with zero false negatives. Feature importance analysis reveals that physics-based kinematic metrics-specifically velocity deviation, GPS distance, and Doppler shift-drive the zero-trust scoring. Ultimately, this framework serves as a proof-of-concept demonstrating that adaptive, low-overhead security integrations are theoretically viable under orbital constraints, paving the way for future physical validation in next-generation 6G networks.