VulnAgent-R2: Evidence-Calibrated Multi-Agent Auditing for Repository-Level Vulnerability Detection
Researchers have developed VulnAgent-R2, an advanced multi-agent auditing framework designed to detect software vulnerabilities at the repository level. This system improves upon previous methods by incorporating modules for counterfactual evidence reweighting, build-aware verification-plan synthesis, and a cost-risk Pareto scheduler. VulnAgent-R2 demonstrates enhanced performance across several benchmark datasets, achieving high F1 and AUROC scores while also reducing the number of computational tokens required for execution. AI
IMPACT Enhances AI's capability in secure software development by improving vulnerability detection accuracy and efficiency.