用Go-Explore测试GPT-4o-mini安全漏洞，发现随机种子影响远超算法设计。
Large Empirical Case Study: Go-Explore adapted for AI Red Team Testing
- 用Go-Explore框架在28次实验中评估安全训练模型的漏洞
- 单次随机种子结果不可靠，多种子平均可降低8倍结果波动
- 奖励塑形导致94%实验探索崩溃，适合做攻击多样性测试
具备工具调用能力的生产级大语言模型虽经安全训练，仍需安全测试。我们针对GPT-4o-mini，在六个研究问题下开展28次实验，应用改进的Go-Explore方法。结果表明，随机种子差异对结果的影响远超算法参数，导致结果差距达8倍；单次种子比较不可靠，而多种子平均显著降低方差。奖励塑形会持续损害性能，在94%的实验中引发探索崩溃，或产生18个无验证攻击的假阳性。在本环境中，简单状态签名优于复杂表示。对于全面安全测试，集成方法可提供攻击类型多样性，而单一代理则更优覆盖特定攻击类型。总体而言，测试安全训练模型时，种子差异与领域知识比算法复杂度更具影响力。
原文摘要 · Abstract (English)
Production LLM agents with tool-using capabilities require security testing despite their safety training. We adapt Go-Explore to evaluate GPT-4o-mini across 28 experimental runs spanning six research questions. We find that random-seed variance dominates algorithmic parameters, yielding an 8x spread in outcomes; single-seed comparisons are unreliable, while multi-seed averaging materially reduces variance in our setup. Reward shaping consistently harms performance, causing exploration collapse in 94% of runs or producing 18 false positives with zero verified attacks. In our environment, simple state signatures outperform complex ones. For comprehensive security testing, ensembles provide attack-type diversity, whereas single agents optimize coverage within a given attack type. Overall, these results suggest that seed variance and targeted domain knowledge can outweigh algorithmic sophistication when testing safety-trained models.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。