Enhancing Automated Penetration Testing Through a Comprehensive Cyber-Security Knowledge Graph

The growing complexity of modern network environments, coupled with threat actors adoption of multi-stage, tool-chain-driven campaigns, has rendered manual penetration testing (pentesting) time-consuming and increasingly ineffective. This study proposes a comprehensive cyber-security knowledge graph (CS-KG) that reconciles disparate data from penetrationtesting tool outputs, the MCP framework, network topology artefacts, asset inventories, and vulnerability databases. By fusing these heterogeneous sources into a high-precision, semantically coherent graph, we enable advanced inference over attack paths, asset criticality, and control efficacy. A reasoning engine built atop the CS-KG dynamically prescribes automated pentesting actions that are context-aware and risk-weighted. Experiments on an enterprise-grade, 1800 -node testbed demonstrate that our framework shortens threat-path discovery time by 63%, raises vulnerability coverage by 48%, and reduces redundant exploit attempts by $\mathbf{4 1 \%}$ relative to state-of-practice automation baselines. The results underline that graph-driven knowledge representation is a tractable route to scalable and intelligent pentesting orchestration for critical infrastructure.