arXiv cs.CL· Zehao Lin, Xixuan Hao, Renyu Fu, Shaobo Cui, Kai Chen, Chunyu Li, Zhiyu Li, Feiyu Xiong·· 14 小时前AI 评分34
LLM 智能体长期记忆安全综述:记忆生命周期内的攻击、防御与治理
A Survey on Long-Term Memory Security in LLM Agents: Attacks, Defenses, and Governance Across the Memory Lifecycle
AI 导读
一篇被 EMNLP 2026 接收的综述提出记忆生命周期框架,沿六个生命周期阶段(写入、存储、检索、执行、共享与传播、遗忘与回滚)和四个安全目标(完整性、机密性、可用性、治理)组织 LLM 智能体长期记忆的攻击、防御及跨阶段依赖。该框架进一步引出可验证记忆治理(VMG),包含五项架构原语,用于在声明的系统与威胁模型边界内指定和评估可审计、可恢复的记忆状态控制。
正文 · 原文
Abstract:The emergence of writable, cross-session persistent memory in LLM agents introduces a qualitatively different threat landscape from conventional input-centric security concerns, characterized by three properties: persistence, statefulness, and propagation. To systematically characterize this landscape, we propose a Memory Lifecycle Framework that organizes attacks, defenses, and their cross-phase dependencies along two axes: six lifecycle phases (Write, Store, Retrieve, Execute, Share & Propagate, Forget & Rollback) and four security objectives (Integrity, Confidentiality, Availability, Governance). This analysis in turn exposes the need for explicit security requirements at the system level, motivating Verifiable Memory Governance (VMG), a framework of five architectural primitives for specifying and evaluating auditable, recoverable control over memory state within declared system and threat-model boundaries. Our analysis motivates integrating provenance, versioning, and policy-aware retention with retrieval and execution defenses to support auditable prevention, containment, and recovery across the lifecycle.
| Comments: | 16 pages, 3 figures, 3 tables. Accepted to EMNLP 2026. Updated references and revised VMG definitions and relationships; revision notes in Appendix F |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Computation and Language (cs.CL) |
| ACM classes: | K.6.5; I.2.0; D.4.6 |
| Cite as: | arXiv:2604.16548 [cs.CR] |
| (or arXiv:2604.16548v4 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2604.16548 arXiv-issued DOI via DataCite |
Submission history
From: Zehao Lin [view email]
[v1]
Fri, 17 Apr 2026 06:28:22 UTC (24,153 KB)
[v2]
Thu, 11 Jun 2026 04:00:49 UTC (517 KB)
[v3]
Tue, 22 Sep 2026 07:33:24 UTC (475 KB)
[v4]
Sat, 24 Oct 2026 05:30:02 UTC (1,559 KB)
来源:arXiv cs.CL · arxiv.org