Graph-Based Quantum Feature Spaces for Cybersecurity Attack Path Inference and Risk Scoring

This paper presents a Graph-Based Quantum Feature Space (GQFS) framework for enterprise cybersecurity that unifies typed attack-graph modeling with quantum-enhanced similarity to improve attack path inference and risk scoring. The method constructs a heterogeneous attack graph from reachability policies, identity relationships, vulnerability intelligence, and telemetry, then applies a quantum feature map to compute kernel-based coherence signals that refine candidate path ranking within a bounded beam-search inference procedure. In empirical evaluation against strong baselines, GQFS achieves Precision@5 = 0.71, Precision@10 = 0.67, Recall@10 = 0.78, and MRR = 0.62, outperforming a graph-only shortest-path baseline (0.52 / 0.49 / 0.57 / 0.44) and a probabilistic edge-likelihood ranking (0.58 / 0.55 / 0.63 / 0.49). Relative to a classical ML scorer without quantum features, GQFS improves early-path relevance (Precision@5: +0.07, MRR: +0.07) while preserving explainability via typed edges and constraint-aware feasibility. An ablation with quantum coherence disabled shows measurable degradation (Precision@10: 0.63 → 0.67, MRR: 0.57 → 0.62), indicating that quantum feature-space similarity contributes primarily to earlier placement of correct paths and reduced “path stitching” inconsistencies. Runtime remains operationally viable for interactive analysis, with median time-to-top10 = 2.09 s (vs 1.18 s graph-only), enabled by limiting quantum-kernel evaluations to candidate expansions rather than exhaustive scoring. Overall, the findings indicate that quantum feature spaces can act as a high-precision refinement layer for attack-path prioritization under enterprise constraints.