arXivDaily arXiv每日学术速递 周一至周五更新
arXiv 2602.15344cs.LG
收藏

ER-MIA:针对长时记忆增强大语言模型的黑盒对抗性记忆注入攻击

ER-MIA: Black-Box Adversarial Memory Injection Attacks on Long-Term Memory-Augmented Large Language Models

Department of Computer Science, The University of Iowa(爱荷华大学计算机科学系);School of Computing, State University of New York at Binghamton(纽约州立大学布法罗分校计算机学院);Department of Computer Science, University of Nevada, Las Vegas(内华达大学拉斯维加斯分校计算机科学系);Department of Computer Science(计算机科学系);Software Engineering, Auburn University(阿伯茨罕大学软件工程系)

Mitchell Piehl, Zhaohan Xi, Zuobin Xiong, Pan He, Muchao Ye

AI总结:

ER-MIA针对长时记忆增强LLM的基于相似性的检索机制提出黑盒对抗性记忆注入攻击,揭示了系统层面的安全漏洞。

摘要:

大语言模型(LLMs)正越来越多地增强长时记忆系统,以克服有限的上下文窗口并使推理在交互中持续进行。然而,近期研究发现,LLMs由于记忆提供了额外的攻击面而变得更加脆弱。在本文中,我们首次系统研究了针对长时记忆增强LLMs中基于相似性的检索机制的黑盒对抗性记忆注入攻击。我们介绍了ER-MIA,一个统一的框架,揭示了这一漏洞,并正式化了两种现实攻击场景:基于内容的攻击和问题定向的攻击。在这些场景中,ER-MIA包括一组可组合的攻击原语和集成攻击,能够在最小的攻击者假设下实现高成功率。在多个LLM和长时记忆系统上的广泛实验表明,基于相似性的检索构成了一个基本且系统层面的漏洞,揭示了在记忆设计和应用场景中持续存在的安全风险。

英文摘要:

Large language models (LLMs) are increasingly augmented with long-term memory systems to overcome finite context windows and enable persistent reasoning across interactions. However, recent research finds that LLMs become more vulnerable because memory provides extra attack surfaces. In this paper, we present the first systematic study of black-box adversarial memory injection attacks that target the similarity-based retrieval mechanism in long-term memory-augmented LLMs. We introduce ER-MIA, a unified framework that exposes this vulnerability and formalizes two realistic attack settings: content-based attacks and question-targeted attacks. In these settings, ER-MIA includes an arsenal of composable attack primitives and ensemble attacks that achieve high success rates under minimal attacker assumptions. Extensive experiments across multiple LLMs and long-term memory systems demonstrate that similarity-based retrieval constitutes a fundamental and system-level vulnerability, revealing security risks that persist across memory designs and application scenarios.

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。

↑