Large Language Models for Vulnerability Detection in Static Code Analysis: A Survey

As software systems grow increasingly complex, security vulnerabilities pose escalating threats to digital infrastructure. This survey examines how Large Language Models (LLMs) enhance static vulnerability detection and systematically categorizes adaptation strategies into five approaches: Base Application, Static Analysis Augmentation, Knowledge Enhancement, Fine-tuning Based, and Hybrid Feature Fusion. Through standardized benchmark comparisons, we demonstrate these strategies' differential impact on detection performance and reveal complementary strengths between encoder-only and decoder-only architectures for varying security priorities. Despite advancements, challenges persist in model robustness, generalization, and adaptation to emerging vulnerabilities. Our comprehensive analysis establishes a foundation for future research while providing practical guidance for selecting and implementing LLM adaptation strategies in security-critical applications.