A Hybrid Post-Quantum Cryptography and Machine Learning and Framework for Intrusion Detection and Downgrade Attack Prevention throughout PQC Migration

Abstract

The evolution from conventional cryptography to post-quantum cryptography (PQC) is underway across managements and innovativeness to alleviate the risk modelled by large-scale quantum processers. Though, throughout this migration era, factual systems remain susceptible to downgrade attacks, where an adversary forces terminuses to transfer weaker, bequest procedures notwithstanding joint PQC support. This study offers a real-world intrusion detection method according to DNNs to classify PQC practice, differentiate legacy traffic, and perceive downgrade attacks among network traffic and bot action. This study proposes a hybrid ML framework-based intrusion detection system (IDS) with a (PQC)-ready protection pipeline. A DT classifier qualified on system flow structures attains great accurateness in distinctive benign from spiteful traffic. Model yields are held using AES-GCM for confidentiality and integrity, with asymmetric key encapsulation (virtual via RSA) and digital signs (virtual via Ed25519) to confirm legitimacy and non-negation. The project is linked: RSA could be substituted by CRYSTALS-Kyber for main encapsulation and Ed25519 by CRYSTALS-Dilithium for signatures short of changing the system architecture. The outcomes prove that the combined ML+PQC pipeline is effective, explainable, and prepared for quantum- tough disposition. We define a label-engineering pipeline which allocates PQC- likeness scores from handshake-derived structures, a downgrade classification approach as per user performance over time, and a class- weighted DNN classifier qualified to discrete PQC, bequest, reduce, and bot programs. Trials on CICIDS2018-derived traffic require test correctness exceptional 98%, with strong performance on the extreme downgrade class. We provide deployment guidance for structure PQC-aware intrusion exposure into actual migration programs.