arXiv cs.AI· Niveen O. Jaffal, Ahmet Yuksel, David Mohaisen·· 10 天前AI 评分22
RAG-PIBench：面向可信 RAG 系统的提示注入检测防泄漏基准
RAG-PIBench: A Leakage-Aware Benchmark for Prompt-Injection Detection in Trustworthy RAG Systems
AI 导读
研究者推出 RAG-PIBench，一个针对 RAG 式提示注入检测的基准，包含 4,876 条上下文样本，划分为冻结的训练、验证与受保护测试集。
正文 · 原文
Abstract: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 competitive. Our results demonstrate the value of leakage-aware benchmark design and strong sparse baselines for reliable prompt-injection detection in RAG systems.
| Comments: | 19 pages, 3 figures, 8 tables |
| Subjects: | Cryptography and Security (cs.CR); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.08571 [cs.CR] |
| (or arXiv:2610.08571v1 [cs.CR] for this version) | |
| https://doi.org/10.48550/arXiv.2610.08571 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: David Mohaisen [view email]
[v1]
Tue, 6 Oct 2026 15:47:47 UTC (162 KB)
来源：arXiv cs.AI · arxiv.org