A Systematic Review of AI-Driven Automated Software Vulnerability Detection Systems

Abstract

This study provides a systematic review of AI-driven software vulnerability detection (AI-SVD) systems published between 2018-2024. Analyzing 26 peer-reviewed papers, it found that C/C++ was the predominant target language, with deep learning models like GNNs and CNNs being widely adopted. Benchmark datasets such as NVD and SARD are extensively used. The review highlights challenges like dataset limitations, language coverage, and model explainability, providing a foundational reference for future security solutions.

Keywords

References