arXiv：cs.AI· Yuming Xu, Mingtao Zhang, Zhuohan Ge, Haoyang Li, Nicole Hu, Yongqi Zhang, Zhiyuan Wen, Jason Chen Zhang, Qing Li, Lei Chen·· 19 小时前
安全检索增强生成（RAG）综述：135 项工作与 SLOT 分类体系
A Survey of Secure Retrieval-Augmented Generation
AI 导读
一篇被 EMNLP 2026 接收的综述系统梳理了检索增强生成（RAG）的安全问题，将安全 RAG 定义为保障外部知识访问安全，并收集了 135 项关于攻击、防御与安全评估的工作。
正文
Abstract:Retrieval-augmented generation (RAG) improves large language models (LLMs) with external knowledge, but this access path creates security risks distinct from inherent prompt-only or parametric-model flaws. We frame secure RAG as securing external knowledge access. We conducted a systematic search and curated 135 works on attacks, defenses, and security evaluation, and organized them with SLOT: a taxonomy along the attack Surface (S) and the corresponding defense Layer (L), with cross-cutting axes Objective (O) and attack-target scope (T). Mapping these studies onto an external knowledge-access pipeline, we expose three mismatches: target mismatch (T2 attacks outpace T2 defenses and evaluation), stage mismatch (S1 attacks outnumber L1 defenses), and signal mismatch (fluent, retrievable, corpus-fitting S1 attacks challenge anomaly-based L2/L3 defenses). Finally, we discuss directions for more realistic targets, surface-complete defense stacks, standardized evaluation, confidentiality, and multimodal and agentic systems, and release the screening process, selected-paper metadata, and machine-readable SLOT labels at this https URL.
| Comments: | Accepted at EMNLP 2026 |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2604.08304 [cs.CR] |
| (or arXiv:2604.08304v5 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2604.08304 arXiv-issued DOI via DataCite |
Submission history
From: Haoyang Li [view email]
[v1]
Thu, 9 Apr 2026 14:38:18 UTC (359 KB)
[v2]
Wed, 27 May 2026 15:22:52 UTC (464 KB)
[v3]
Mon, 8 Jun 2026 14:08:37 UTC (465 KB)
[v4]
Wed, 7 Oct 2026 13:12:33 UTC (473 KB)
[v5]
Mon, 12 Oct 2026 14:48:56 UTC (473 KB)
来源：arXiv：cs.AI · arxiv.org