Sifting the Noise: A Comparative Study of LLM Agents in Vulnerability False Positive Filtering

TL;DR

This study compares three LLM-based agent frameworks for filtering false positives in static security testing, showing they can significantly reduce noise but with varying effectiveness depending on model strength and vulnerability type.

Contribution

It provides a comprehensive comparison of LLM agent architectures for vulnerability false positive filtering, highlighting their strengths, limitations, and cost considerations.

Findings

LLM agents can reduce false positives from over 92% to as low as 6.3%.

Performance varies significantly with backbone model strength and vulnerability type.

Aggressive false positive reduction may suppress true vulnerabilities.

Abstract

Static Application Security Testing (SAST) tools are essential for identifying software vulnerabilities, but they often produce a high volume of false positives (FPs), imposing a substantial manual triage burden on developers. Recent advances in Large Language Model (LLM) agents offer a promising direction by enabling iterative reasoning, tool use, and environment interaction to refine SAST alerts. However, the comparative effectiveness of different LLM-based agent architectures for FP filtering remains poorly understood. In this paper, we present a comparative study of three state-of-the-art LLM-based agent frameworks, i.e., Aider, OpenHands, and SWE-agent, for vulnerability FP filtering. We evaluate these frameworks using the vulnerabilities from the OWASP Benchmark and real-world open-source Java projects. The experimental results show that LLM-based agents can remove the majority of SAST noise, reducing an initial FP detection rate of over 92% on the OWASP Benchmark to as low as 6.3% in the best configuration. On real-world dataset, the best configuration of LLM-based agents can achieve an FP identification rate of up to 93.3% involving CodeQL alerts. However, the benefits of agents are strongly backbone- and CWE-dependent: agentic frameworks significantly outperform vanilla prompting for stronger models such as Claude Sonnet 4 and GPT-5, but yield limited or inconsistent gains for weaker backbones. Moreover, aggressive FP reduction can come at the cost of suppressing true vulnerabilities, highlighting important trade-offs. Finally, we observe large disparities in computational cost across agent frameworks. Overall, our study demonstrates that LLM-based agents are a powerful but non-uniform solution for SAST FP filtering, and that their practical deployment requires careful consideration of agent design, backbone model choice, vulnerability category, and operational cost.