Quantum-Resistant AI Models for Intrusion Detection

Abstract

As quantum computing transitions from theoretical constructs to practical implementations, classical cryptographic foundations—long the bedrock of cybersecurity—face unprecedented vulnerability. Shor’s algorithm and related quantum techniques threaten to render public-key schemes like RSA and ECC obsolete, exposing sensitive communications and stored data to adversarial compromise. In this context, Intrusion Detection Systems (IDS) that harness artificial intelligence (AI) must evolve to incorporate quantum-resistant mechanisms to secure both their operational logic and the data they process. This manuscript presents an end-to-end framework for a quantum-resistant AI-based IDS, integrating lattice-based public-key encryption (FrodoKEM) for feature confidentiality and Cheon-Kim-Kim-Song (CKKS) homomorphic encryption for privacy-preserving inference. We develop and optimize a deep neural network with encrypted weights and encrypted input vectors, enabling packet inspection and anomaly detection entirely in the encrypted domain. Experiments on UNSW-NB15 and CIC-IDS2017 datasets demonstrate that our quantum-resistant IDS achieves an average detection accuracy of 97.2%, a true positive rate of 96.0%, and a false positive rate of 3.5%, closely matching plaintext performance while incurring a total latency overhead of approximately 25% (14.8 ms encrypted inference vs. 12.1 ms plaintext, plus 2.9 ms encryption/decryption). We analyze computational and memory trade-offs, explore key-management strategies compliant with NIST PQC guidelines, and discuss scalability considerations for real-world deployment. Our findings confirm that integrating post-quantum cryptography into AI-driven IDS can future-proof network security infrastructures against both classical and quantum adversaries, with acceptable performance overhead for enterprise environments.