arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2412.05232cs.CL
收藏

LIAR:利用推理时对齐(Best-of-N)在数秒内越狱大语言模型

LIAR: Leveraging Inference Time Alignment (Best-of-N) to Jailbreak LLMs in Seconds

University of Central Florida(中央佛罗里达大学);University of Maryland, College Park(马里兰大学 College Park 分校);Princeton University(普林斯顿大学);Capital One(Capital One 公司)

James Beetham, Souradip Chakraborty, Mengdi Wang, Furong Huang, Amrit Singh Bedi, Mubarak Shah

AI总结:

本文提出 LIAR,一种无需训练的黑盒 Best-of-N 采样越狱攻击方法,将越狱重新定义为推理时失对齐问题,在保持顶尖成功率的同时将攻击时间从小时级降至秒级,并引入理论指标量化安全对齐强度。

摘要:

越狱攻击通过精心构造的提示词诱发有害输出,暴露了安全对齐大语言模型的脆弱性。现有方法依赖离散优化或训练对抗性生成器,但速度慢、计算密集且往往不切实际。我们认为这些低效源于对问题的错误刻画。相反,我们将越狱重新定义为推理时失对齐,并提出 LIAR(Leveraging Inference-time misAlignment to jailbReak,利用推理时失对齐进行越狱),这是一种快速、黑盒、Best-of-N 采样攻击,无需训练。LIAR 达到了与最先进方法相当的成功率,同时将困惑度降低 10 倍,并将攻击时间(Time-to-Attack)从数小时缩短至数秒。我们还引入了一个理论上的“防越狱安全网”指标,用于量化安全对齐强度,并推导出次优性界限。我们的工作为评估大语言模型鲁棒性和推进对齐研究提供了一个简单而有效的工具。

英文摘要:

Jailbreak attacks expose vulnerabilities in safety-aligned LLMs by eliciting harmful outputs through carefully crafted prompts. Existing methods rely on discrete optimization or trained adversarial generators, but are slow, compute-intensive, and often impractical. We argue that these inefficiencies stem from a mischaracterization of the problem. Instead, we frame jailbreaks as inference-time misalignment and introduce LIAR (Leveraging Inference-time misAlignment to jailbReak), a fast, black-box, best-of-$N$ sampling attack requiring no training. LIAR matches state-of-the-art success rates while reducing perplexity by $10\times$ and Time-to-Attack from hours to seconds. We also introduce a theoretical "safety net against jailbreaks" metric to quantify safety alignment strength and derive suboptimality bounds. Our work offers a simple yet effective tool for evaluating LLM robustness and advancing alignment research.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑