Researchers have questioned the effectiveness of quantum machine learning (QML) in network intrusion detection systems. A new study compares QML models with classical models and finds that well-tuned classical models remain competitive. The study uses a unified benchmark to evaluate hybrid variational quantum circuits, quantum-kernel SVMs, and five classical baselines across four standard datasets. Results show that tuned classical models match or exceed the performance of QML models on aggregate detection, and an attribution audit attributes this to classical preprocessing and regularisation rather than quantum effects. However, two advantages of QML are found: a quantum-kernel SVM outperforms its direct classical surrogate in certain metrics, and a small four-qubit hybrid model out-detects the best classical baseline in one task. --- Why it matters: This study matters to engineers and researchers because it challenges the assumption that QML is superior to classical machine learning for network intrusion detection. The results have implications for the development of more efficient and effective security systems, and highlight the need for rigorous evaluation and comparison of different approaches. Source: https://arxiv.org/abs/2608.18155