PQ-IDPS: Adversarially Robust Intrusion Detection for Post-Quantum Encrypted Traffic with Hybrid Classical-Quantum Machine Learning

The imminent deployment of post-quantum cryptography standards fundamentally transforms encrypted network traffic characteristics, rendering existing intrusion detection systems ineffective against traffic patterns generated by latticebased key encapsulation mechanisms and hash-based digital signatures. Although the National Institute of Standards and Technology finalized post-quantum cryptographic algorithms in August 2024, including ML-KEM based on CRYSTALS-Kyber and MLDSA based on CRYSTALS-Dilithium, contemporary intrusion detection approaches remain exclusively trained on classical TLS handshakes, creating critical security vulnerabilities during the cryptographic transition period. This paper introduces PQIDPS, a hybrid classical-quantum machine learning framework specifically designed for adversarially robust intrusion detection in post-quantum encrypted traffic environments. Our approach combines convolutional-LSTM architectures optimized for classical TLS with variational quantum classifiers that leverage quantum parallelism for post-quantum protocol analysis, unified through an adaptive ensemble mechanism that automatically detects protocol types and weights component contributions. To address the emerging threat of quantum-enhanced adversarial attacks, we develop a comprehensive defense framework incorporating Lipschitz-bounded network architectures, quantum noise injection during training, and randomized smoothing certification that provides provable robustness bounds against Grover-optimized perturbations. Through extensive evaluation on three novel datasets comprising CESNET-TLS-22 hybrid traffic, custom QUIC-PQC traces with pure Kyber key exchanges, and IoTPQC captures featuring Dilithium signatures, PQ-IDPS achieves 95.3% detection accuracy on hybrid classical-PQC traffic while maintaining certified robustness against quantum adversarial attacks bounded by O( √ N) Grover advantage. Our comprehensive ablation studies demonstrate that the quantum component contributes 7.8% accuracy improvement on pure post-quantum traffic compared to classical-only baselines, while hybrid ensemble fusion achieves 98.1% accuracy on protocol-mixed environments. The framework provides practical deployment guidelines for network operators transitioning to post-quantum cryptography, including traffic characterization tools, adversarial robustness benchmarks, and NIST migration recommendations. This work establishes the first comprehensive approach to adversarially robust intrusion detection for the post-quantum era, addressing both cryptographic modernization and quantum threat mitigation in a unified framework.