两篇新研究论文探讨了大型语言模型（LLM）代理内存系统的漏洞。其中一篇论文 MemPoison 详细介绍了一种隐蔽攻击，该攻击通过对话将可触发的后门注入代理的长期记忆中，成功地以高达 0.95 的成功率误导其未来响应。另一篇论文 MRMMIA 介绍了一种专门针对聊天代理内存的成员推理攻击方法，通过推断特定数据单元是否属于代理的内存存储，展示了重大的隐私风险。 AI
影响 这些发现突显了 LLM 代理中关键的安全和隐私漏洞，可能影响用户信任并需要新的防御机制。
排序理由 两篇在 arXiv 上发表的学术论文，详细介绍了 LLM 代理内存系统中的新型攻击向量和隐私风险。
在 Hugging Face Daily Papers 阅读 →
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
arXiv:2605.29960v1 Announce Type: cross Abstract: Large language model (LLM) agents increasingly leverage long term memory to support persistent and autonomous task execution. However, this capability also introduces a new attack surface: memory poisoning, where adversaries can i…
arXiv:2605.27825v1 Announce Type: cross Abstract: Membership inference attacks (MIAs) test whether a target data record belongs to a system's private data, and have become a standard tool to measure privacy leakage in machine learning systems. Prior work has primarily focused on …
Membership inference attacks (MIAs) test whether a target data record belongs to a system's private data, and have become a standard tool to measure privacy leakage in machine learning systems. Prior work has primarily focused on training corpora or retrieval databases. However, …