Multi-Agent LLM System for Automated Vulnerability Discovery and Reproduction

FuzzingBrain V2 is a multi-agent LLM system that automates vulnerability discovery and reproduction by combining fuzzing with large language models to achieve a 90% detection rate on benchmark datasets and discovered 29 zero-day vulnerabilities in real-world testing. The system addresses key limitations of existing approaches through innovations including fuzzer-reproducible vulnerability verification, control-flow-based localization, hierarchical function analysis, and enhanced context engineering for complex vulnerability reasoning.