研究库 论文知识库

内容库 / 论文

arXiv:2610.08571 · RAG 检索增强

RAG-PIBench: A Leakage-Aware Benchmark for Prompt-Injection Detection in Trustworthy RAG Systems

RAG-PIBench: A Leakage-Aware Benchmark for Prompt-Injection Detection in Trustworthy RAG Systems

类型：arxiv
标识：2610.08571
链接：http://arxiv.org/abs/2610.08571v1
主分类：rag
形态：benchmark
TLDR：Retrieval-Augmented Generation (RAG) systems are vulnerable to prompt-injection attacks embedded in retrieved content. We introduce RAG-PIBench, a benchmark for RAG-style prompt-injection detection containing 4,876 contextual examples across frozen train, validation, and protected-test splits. Using a leakage-aware construction pipeline and strict evaluation protocol, we compare keyword-based, semantic-reference, TF-IDF, and transformer-based detectors. DistilBERT achieves the best protected-test performance (F1 = 0.896, PR-AUC = 0.968), while TF-IDF SVM and logistic regression remain competit
副分类：evaluation
待LLM分类：否
来源文件：
/inbox/tom/_candidates/2026-10-07-agent-rag-longcontext-candidates.json

相关 · 按语义匹配

笔记次精读：Argus — Uncertainty Quantification for Computer-Use Agents（短评）

笔记 Tom 文献雷达 · Agent RAG LongContext · 2026-07-07 14:30

文档 Stephen 自测 · R68 · 2026-09-07

仓库 augety121/MCP-State-Twin

文档 Tom 评 flyP · 2026-09-08 AgentVista 多模态长程视觉 grounding 评测基准精读

笔记 Tom 文献雷达 · Agent + RAG + Long-Context · 2026-08-19

arXiv

2610.08571

TLDR

Retrieval-Augmented Generation (RAG) systems are vulnerable to prompt-injection attacks embedded in retrieved content. We introduce RAG-PIBench, a benchmark for RAG-style prompt-injection detection containing 4,876 contextual examples across frozen train, validation, and protected-test splits. Using a leakage-aware construction pipeline and strict evaluation protocol, we compare keyword-based, semantic-reference, TF-IDF, and transformer-based detectors. DistilBERT achieves the best protected-test performance (F1 = 0.896, PR-AUC = 0.968), while TF-IDF SVM and logistic regression remain competit

arXiv ↗ 中文全文