Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agents

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Zhan, Qiusi, Fang, Richard, Panchal, Henil Shalin, Kang, Daniel
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917944441700352
author Zhan, Qiusi
Fang, Richard
Panchal, Henil Shalin
Kang, Daniel
author_facet Zhan, Qiusi
Fang, Richard
Panchal, Henil Shalin
Kang, Daniel
contents Large Language Model (LLM) agents exhibit remarkable performance across diverse applications by using external tools to interact with environments. However, integrating external tools introduces security risks, such as indirect prompt injection (IPI) attacks. Despite defenses designed for IPI attacks, their robustness remains questionable due to insufficient testing against adaptive attacks. In this paper, we evaluate eight different defenses and bypass all of them using adaptive attacks, consistently achieving an attack success rate of over 50%. This reveals critical vulnerabilities in current defenses. Our research underscores the need for adaptive attack evaluation when designing defenses to ensure robustness and reliability. The code is available at https://github.com/uiuc-kang-lab/AdaptiveAttackAgent.
format Preprint
id arxiv_https___arxiv_org_abs_2503_00061
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agents
Zhan, Qiusi
Fang, Richard
Panchal, Henil Shalin
Kang, Daniel
Cryptography and Security
Machine Learning
Large Language Model (LLM) agents exhibit remarkable performance across diverse applications by using external tools to interact with environments. However, integrating external tools introduces security risks, such as indirect prompt injection (IPI) attacks. Despite defenses designed for IPI attacks, their robustness remains questionable due to insufficient testing against adaptive attacks. In this paper, we evaluate eight different defenses and bypass all of them using adaptive attacks, consistently achieving an attack success rate of over 50%. This reveals critical vulnerabilities in current defenses. Our research underscores the need for adaptive attack evaluation when designing defenses to ensure robustness and reliability. The code is available at https://github.com/uiuc-kang-lab/AdaptiveAttackAgent.
title Adaptive Attacks Break Defenses Against Indirect Prompt Injection Attacks on LLM Agents
topic Cryptography and Security
Machine Learning
url https://arxiv.org/abs/2503.00061