Defending against Indirect Prompt Injection by Instruction Detection

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Wen, Tongyu, Wang, Chenglong, Yang, Xiyuan, Tang, Haoyu, Xie, Yueqi, Lyu, Lingjuan, Dou, Zhicheng, Wu, Fangzhao
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918273561395200
author Wen, Tongyu
Wang, Chenglong
Yang, Xiyuan
Tang, Haoyu
Xie, Yueqi
Lyu, Lingjuan
Dou, Zhicheng
Wu, Fangzhao
author_facet Wen, Tongyu
Wang, Chenglong
Yang, Xiyuan
Tang, Haoyu
Xie, Yueqi
Lyu, Lingjuan
Dou, Zhicheng
Wu, Fangzhao
contents The integration of Large Language Models (LLMs) with external sources is becoming increasingly common, with Retrieval-Augmented Generation (RAG) being a prominent example. However, this integration introduces vulnerabilities of Indirect Prompt Injection (IPI) attacks, where hidden instructions embedded in external data can manipulate LLMs into executing unintended or harmful actions. We recognize that IPI attacks fundamentally rely on the presence of instructions embedded within external content, which can alter the behavioral states of LLMs. Can the effective detection of such state changes help us defend against IPI attacks? In this paper, we propose InstructDetector, a novel detection-based approach that leverages the behavioral states of LLMs to identify potential IPI attacks. Specifically, we demonstrate the hidden states and gradients from intermediate layers provide highly discriminative features for instruction detection. By effectively combining these features, InstructDetector achieves a detection accuracy of 99.60% in the in-domain setting and 96.90% in the out-of-domain setting, and reduces the attack success rate to just 0.03% on the BIPIA benchmark. The code is publicly available at https://github.com/MYVAE/Instruction-detection.
format Preprint
id arxiv_https___arxiv_org_abs_2505_06311
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Defending against Indirect Prompt Injection by Instruction Detection
Wen, Tongyu
Wang, Chenglong
Yang, Xiyuan
Tang, Haoyu
Xie, Yueqi
Lyu, Lingjuan
Dou, Zhicheng
Wu, Fangzhao
Cryptography and Security
Artificial Intelligence
The integration of Large Language Models (LLMs) with external sources is becoming increasingly common, with Retrieval-Augmented Generation (RAG) being a prominent example. However, this integration introduces vulnerabilities of Indirect Prompt Injection (IPI) attacks, where hidden instructions embedded in external data can manipulate LLMs into executing unintended or harmful actions. We recognize that IPI attacks fundamentally rely on the presence of instructions embedded within external content, which can alter the behavioral states of LLMs. Can the effective detection of such state changes help us defend against IPI attacks? In this paper, we propose InstructDetector, a novel detection-based approach that leverages the behavioral states of LLMs to identify potential IPI attacks. Specifically, we demonstrate the hidden states and gradients from intermediate layers provide highly discriminative features for instruction detection. By effectively combining these features, InstructDetector achieves a detection accuracy of 99.60% in the in-domain setting and 96.90% in the out-of-domain setting, and reduces the attack success rate to just 0.03% on the BIPIA benchmark. The code is publicly available at https://github.com/MYVAE/Instruction-detection.
title Defending against Indirect Prompt Injection by Instruction Detection
topic Cryptography and Security
Artificial Intelligence
url https://arxiv.org/abs/2505.06311