Combining LLM and Dynamic Analysis for Universal Testing of Static Application Security Testing Tools
LLM4SAST is a universal testing framework that uses large language models to find faults in static application security testing (SAST) tools across different programming languages. It generates diverse program variants using LLMs guided by metamorphic relations and filters them using dynamic feedback and clustering. The system successfully found 66 unique faults, significantly outperforming existing methods by finding more bugs with fewer test variants.