Quantum kernel methods enable the evaluation of kernel functions in exponentially large Hilbert spaces with near-term quantum hardware, offering a pragmatic route to quantum-enhanced machine learning. This survey synthesises recent progress, focusing on demonstrators in cybersecurity –network intrusion detection, fraud screening, and visual inspection– where shallow quantum circuits have attained parity or modest gains over optimised classical baselines. Common design choices are identified, including hybrid quantum–classical workflows and noise-aware, problem-specific feature maps. Persistent challenges, including limited qubit availability, decoherence effects, sampling overhead, and data governance constraints, are thoroughly examined. The survey concludes by outlining research priorities: automated feature-map discovery, rigorous benchmarking against state-of-the-art classical learners, and integration strategies compatible with contemporary security pipelines. The analysis is intended to assist researchers and practitioners in determining when quantum kernels are likely to yield a measurable advantage on Noisy Intermediate-Scale Quantum devices (NISQ).