GAT-APG: Graph Attention Network-Based Attack Path Generation for Security Simulation

Existing attack path generation methods face limitations in dynamic simulation environments due to their reliance on static network models and computational inefficiencies when network configurations change frequently. This study proposes GAT-APG, a reinforcement learning framework that combines Graph Attention Networks with policy gradient methods to generate adaptive attack paths for security simulation. The approach employs node-centric vulnerability assessment that transforms network traffic data into vulnerability metrics, enabling adaptation to network changes without complete graph recalculation. Experimental validation on controlled graphs and the Kyoto dataset demonstrates competitive performance, achieving 93% accuracy against brute-force methods and showing 69.1% exact matches with Dijkstra’s algorithm on real-world topologies. The framework provides a simulation-ready environment for vulnerability assessment and defensive planning in dynamic networks.