Machine Learning-Based Vulnerability Detection in AI-Generated Source Code: A Systematic Literature Review and Research Directions

Abstract

The rapid adoption of Generative AI and Large Language Models (LLMs) has transformed software development, but AI-generated code may contain security vulnerabilities such as injection flaws, weak authentication, and unsafe API usage. Traditional rule-based and static-analysis techniques struggle with semantically diverse generated code. This paper presents a systematic literature review of Machine Learning (ML), Deep Learning (DL), Graph Neural Network, Transformer, and LLM-based vulnerability detection, with emphasis on AI-generated source code. It compares approaches, datasets, and code representations, and examines challenges including data imbalance, false positives, limited AI-generated datasets, and explainability. A conceptual framework combining preprocessing, structural and semantic representation, ML/DL classification, CWE categorization, explainability, and validation is proposed. Future directions include hybrid ML-LLM models, real-time IDE integration, and automated vulnerability repair..