---
id: 20260311-T0-18
title: "FuzzingRL 强化模糊测试方法发布"
title_en: "FuzzingRL Enhances Fuzz Testing Method"
url: https://ai.daily.yangsir.net/daily/20260311-T0-18
issue_date: 2026-03-11
publish_date: 2026-03-10T04:00:00.000Z
source_name: "arXiv cs.LG (ML)"
source_url: https://arxiv.org/abs/2603.06600
---

# FuzzingRL 强化模糊测试方法发布

arXiv 论文提出 FuzzingRL 方法，通过强化学习自动生成测试问题以暴露视觉语言模型（VLM）的缺陷。该方法在 20 个 VLM 上测试，发现 1000+ 个边界案例，准确率达 88%。相比传统方法，测试效率提升 3 倍，可应用于自动驾驶和医疗影像等安全敏感领域，帮助开发者提前发现系统漏洞。

## English Version

**FuzzingRL Enhances Fuzz Testing Method**

arXiv paper proposes FuzzingRL method, using reinforcement learning to auto-generate test cases exposing vision-language model (VLM) vulnerabilities. Tested on 20 VLMs, it discovered 1000+ edge cases with 88% accuracy. Traditional methods are 3x slower, and FuzzingRL applies to safety-critical domains like autonomous driving and medical imaging, helping developers proactively identify system vulnerabilities.

---

**来源**：[arXiv cs.LG (ML)](https://arxiv.org/abs/2603.06600)

**详情页**：https://ai.daily.yangsir.net/daily/20260311-T0-18

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*