Graph Neural Networks for Interprocedural Vulnerability Detection Using Software Dependency and Program-Flow Graphs

Authors

  • Talha Hussain Faculty of Computing, Riphah International University, Islamabad 46000, Pakistan Author
  • Farhan Bashir Shaikh Assistant Professor, Department of Computer Science, The University of Larkano Author

DOI:

https://doi.org/10.71317/kjard.2.11.2026.1093

Keywords:

Graph Neural Networks, Vulnerability Detection, Interprocedural Analysis, Code Property Graph, System Dependence Graph, Program Slicing, Taint Analysis, Static Analysis, Deep Learning, Software Security

Abstract

Software vulnerabilities remain a critical threat to modern computing infrastructure, with reported flaws increasing annually and attack surfaces expanding rapidly through open-source ecosystems. Traditional detection approaches static rule-based analyzers, dynamic fuzzing, and symbolic execution suffer from high false-positive rates, path explosion, and poor scalability. Early machine learning methods treating code as flat token sequences fail to capture structural and semantic dependencies, experiencing severe performance degradation on realistic benchmarks. Graph Neural Networks (GNNs) have emerged as a powerful paradigm, modeling programs as structured graphs including Abstract Syntax Trees, Control Flow Graphs, Data Flow Graphs, and Program Dependence Graphs. However, most existing GNN-based detectors operate intraprocedurally, missing cross-function vulnerabilities such as buffer overflows and use-after-free that depend on interprocedural data flows. This review provides a comprehensive evaluation of GNN architectures, software dependency graph representations, and program-flow models for interprocedural vulnerability detection. We examine theoretical foundations of Code Property Graphs and System Dependence Graphs, analyze advanced architectures including relational, hyperbolic, quantum-enhanced, and hybrid language-graph models, and explore interprocedural flow mechanics incorporating program slicing, taint analysis, and executable vulnerability chains. We present comparative analyses of state-of-the-art detectors, evaluate benchmark datasets and label quality issues, and investigate explainability, causal graph learning, and adversarial robustness. Finally, we identify open challenges including graph construction scalability, long-range dependency modeling, LLM integration, and automated patch generation, synthesizing key conclusions to guide future research in scalable, interpretable, and robust vulnerability detection.

Downloads

Published

2026-11-21

How to Cite

Hussain, T., & Shaikh, F. B. (2026). Graph Neural Networks for Interprocedural Vulnerability Detection Using Software Dependency and Program-Flow Graphs. Kashmir Journal of Academic Research and Development, 2(11), 409-422. https://doi.org/10.71317/kjard.2.11.2026.1093