Converged Security Architectures for Critical Infrastructure: Post-Quantum Cryptography, Graph Neural Intrusion Detection and Cross-Domain Empirical Validation

Abstract

This paper introduces a unified, three-layer converged security architecture designed to shield critical digital infrastructure from both classical cyber threats and emerging quantum decryption risks. The architecture integrates a DNA-steganography post-quantum key exchange scheme to secure inter-layer communications, a temporal graph residual network optimized to detect structural anomalies on communication buses (like UAV and industrial CAN networks), and a classical machine learning threat triage system using transfer learning with multi-source SIEM risk scoring. Furthermore, the authors conduct a cross-domain transfer analysis, demonstrating that hybrid quantum-classical classifiers built for quishing detection successfully generalize to offline signature forgery detection. Evaluation on public intrusion detection benchmarks (CIC-IDS2018 and UNSW-NB15) shows that this unified approach maintains state-of-the-art detection accuracy while reducing mean data preprocessing and detection latency by 28% compared to deploying the components independently.