arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2603.11149cs.LGcs.CR
收藏

大语言模型中越狱攻击的系统缩放分析

Systematic Scaling Analysis of Jailbreak Attacks in Large Language Models

University of Illinois Urbana-Champaign(伊利诺伊大学厄巴纳-香槟分校);Google DeepMind(谷歌DeepMind)

Xiangwen Wang, Ananth Balashankar, Varun Chandrasekaran

AI总结:

本文系统分析了大语言模型中越狱攻击的缩放规律,通过计算限制优化过程评估不同方法、模型家族和危害类型的攻击成功率,发现提示基方法在计算效率上更具优势。

摘要:

大语言模型仍然容易受到越狱攻击的影响,但缺乏对攻击成功率如何随着攻击者努力程度缩放的系统理解。我们通过将每次攻击视为计算受限的优化过程,并在共享的FLOPs轴上测量进展,建立了越狱的缩放定律框架。我们的系统评估涵盖了四种代表性的越狱范式,包括基于优化的攻击、自我完善提示、基于采样的选择以及遗传优化,覆盖了多种模型家族和规模上的多样化有害目标。我们研究了将攻击预算与攻击成功率相关联的缩放定律,通过拟合简单的饱和指数函数到FLOPs-成功率轨迹上,并从拟合曲线中推导出可比的效率总结。经验上,基于提示的范式相比基于优化的方法更有效率。为了解释这一差距,我们将基于提示的更新转换为优化视角,并通过相同状态的比较表明,基于提示的攻击更有效地在提示空间中进行优化。我们还发现,基于提示的方法占据着高成功率、高隐蔽性的操作点。最后,我们发现漏洞与目标密切相关:涉及虚假信息的伤害通常比其他非虚假信息伤害更容易引发。

英文摘要:

Large language models remain vulnerable to jailbreak attacks, yet we still lack a systematic understanding of how jailbreak success scales with attacker effort across methods, model families, and harm types. We initiate a scaling-law framework for jailbreaks by treating each attack as a compute-bounded optimization procedure and measuring progress on a shared FLOPs axis. Our systematic evaluation spans four representative jailbreak paradigms, covering optimization-based attacks, self-refinement prompting, sampling-based selection, and genetic optimization, across multiple model families and scales on a diverse set of harmful goals. We investigate scaling laws that relate attacker budget to attack success score by fitting a simple saturating exponential function to FLOPs--success trajectories, and we derive comparable efficiency summaries from the fitted curves. Empirically, prompting-based paradigms tend to be the most compute-efficient compared to optimization-based methods. To explain this gap, we cast prompt-based updates into an optimization view and show via a same-state comparison that prompt-based attacks more effectively optimize in prompt space. We also show that attacks occupy distinct success--stealthiness operating points with prompting-based methods occupying the high-success, high-stealth region. Finally, we find that vulnerability is strongly goal-dependent: harms involving misinformation are typically easier to elicit than other non-misinformation harms.

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。

↑