Enregistré dans:
Détails bibliographiques
Auteurs principaux: Yao, Ruobing, Zhang, Yifei, Song, Shuang, Gao, Neng, Tu, Chenyang
Format: Preprint
Publié: 2025
Sujets:
Accès en ligne:https://arxiv.org/abs/2505.13506
Tags: Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
_version_ 1866915293249404928
author Yao, Ruobing
Zhang, Yifei
Song, Shuang
Gao, Neng
Tu, Chenyang
author_facet Yao, Ruobing
Zhang, Yifei
Song, Shuang
Gao, Neng
Tu, Chenyang
contents Retrieval-Augmented Generation (RAG) compensates for the static knowledge limitations of Large Language Models (LLMs) by integrating external knowledge, producing responses with enhanced factual correctness and query-specific contextualization. However, it also introduces new attack surfaces such as corpus poisoning at the same time. Most of the existing defense methods rely on the internal knowledge of the model, which conflicts with the design concept of RAG. To bridge the gap, EcoSafeRAG uses sentence-level processing and bait-guided context diversity detection to identify malicious content by analyzing the context diversity of candidate documents without relying on LLM internal knowledge. Experiments show EcoSafeRAG delivers state-of-the-art security with plug-and-play deployment, simultaneously improving clean-scenario RAG performance while maintaining practical operational costs (relatively 1.2$\times$ latency, 48\%-80\% token reduction versus Vanilla RAG).
format Preprint
id arxiv_https___arxiv_org_abs_2505_13506
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle EcoSafeRAG: Efficient Security through Context Analysis in Retrieval-Augmented Generation
Yao, Ruobing
Zhang, Yifei
Song, Shuang
Gao, Neng
Tu, Chenyang
Computation and Language
Artificial Intelligence
Retrieval-Augmented Generation (RAG) compensates for the static knowledge limitations of Large Language Models (LLMs) by integrating external knowledge, producing responses with enhanced factual correctness and query-specific contextualization. However, it also introduces new attack surfaces such as corpus poisoning at the same time. Most of the existing defense methods rely on the internal knowledge of the model, which conflicts with the design concept of RAG. To bridge the gap, EcoSafeRAG uses sentence-level processing and bait-guided context diversity detection to identify malicious content by analyzing the context diversity of candidate documents without relying on LLM internal knowledge. Experiments show EcoSafeRAG delivers state-of-the-art security with plug-and-play deployment, simultaneously improving clean-scenario RAG performance while maintaining practical operational costs (relatively 1.2$\times$ latency, 48\%-80\% token reduction versus Vanilla RAG).
title EcoSafeRAG: Efficient Security through Context Analysis in Retrieval-Augmented Generation
topic Computation and Language
Artificial Intelligence
url https://arxiv.org/abs/2505.13506