两篇新研究论文MemMorph和Evo-Attacker详细介绍了通过利用大型语言模型（LLM）代理的内存系统进行攻击的新方法。MemMorph通过精心设计的记录来毒害代理的长期记忆，以微妙地影响工具选择，用少量数据即可实现高成功率。Evo-Attacker使用内存增强强化学习来动态演化针对基于LLM的多代理系统（LLM-MAS）中长时域工具操纵的攻击策略。两项研究都强调了LLM代理内存模块的脆弱性，并强调了对健壮的内存级安全防护措施的迫切需求。 AI
影响 这些研究揭示了LLM代理内存系统的关键漏洞，需要开发新的安全措施来防止恶意工具劫持。
排序理由 两篇在arXiv上发表的学术论文，详细介绍了针对LLM代理的新型攻击向量。
在 arXiv cs.MA (Multiagent) 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2605.26154v1 Announce Type: cross Abstract: LLM-driven agents are capable of selecting external tools to complete users' tasks. However, attackers could compromise such process, steering agents toward inappropriate/wrong tools and enabling malicious actions. Most existing a…
arXiv:2605.25389v1 Announce Type: cross Abstract: While Large Language Model-based Multi-Agent Systems (LLM-MAS) demonstrate remarkable capabilities in solving complex tasks by orchestrating specialized agents and external tools, the implicit trust in tool outputs creates a criti…
While Large Language Model-based Multi-Agent Systems (LLM-MAS) demonstrate remarkable capabilities in solving complex tasks by orchestrating specialized agents and external tools, the implicit trust in tool outputs creates a critical attack surface. Existing tool attacks are limi…