Quantum-Inspired Cyber Threat Detection and Adaptive Security Management for IoT Networks

Abstract

The Internet of Things (IoT) and resource-constrained devices are becoming ubiquitous in smart homes, industrial control systems, and critical infrastructure, creating a large attack surface that is difficult to characterize in real-time with traditional signatures and classical machine learning classifiers. This paper introduces a Quantum Inspired Cyber Threat Detection and Adaptive Security (QI-CTAS) framework that leverages the mathematical models from quantum computing used for superposed features states, amplitude-weighted probability, and entanglement-inspired correlation, and applies the models to model non-linear, high-dimensional cyber threat interactions that Euclidean feature spaces fail to capture. QI-CTAS is made up of four cooperating layers, including quantum-inspired feature encoding, an amplitude-based threat scoring engine, adaptive security orchestration, and a continuous device-trust learning layer, all of which are in a closed loop that adjusts the degree of the device's trustworthiness based on network conditions. QI-CTAS outperforms the best classical baseline that is a tuned ensemble of Random Forest classifiers by 7.5, 8.5, and 6.6 percentage points for detection accuracy, F1-score, and false positive rate, respectively, after processing more than 14 months of labelled network flow data collected from a composite testbed consisting of 46 representative IoT device classes deployed in three environments: a smart building, an industrial IoT gateway cluster, and a consumer mesh network. By removing each of the quantum-inspired components one by one from the ablation experiments, it is found that the majority of the accuracy gain is achieved by removing Grover-inspired amplitude amplification and entanglement-based cross-device correlation, with an F1-score drop of 0.059 and 0.077, respectively. The average time between anomaly detection and the first adaptive policy action is 3.4 minutes, which is 89% less than a static rule-based baseline for intrusion detection. These results show that quantum-inspired representations, even in the absence of physical quantum hardware, provide a computationally practical and empirically proven approach towards low-latency, adaptive threat detection in heterogeneous ecosystems of IoT devices.