Mimicking the Familiar: Dynamic Command Generation for Information Theft Attacks in LLM Tool-Learning System

TL;DR

This paper introduces AutoCMD, a dynamic command generation method that mimics familiar patterns to improve information theft attacks on LLM tool-learning systems, revealing vulnerabilities and proposing defenses.

Contribution

AutoCMD is a novel approach that learns to generate targeted malicious commands by mimicking upstream tool information, enhancing attack success rates in dynamic environments.

Findings

AutoCMD achieves +13.2% increase in attack success rate.

AutoCMD generalizes to new tool-learning systems.

Four defense methods effectively mitigate the attack.

Abstract

Information theft attacks pose a significant risk to Large Language Model (LLM) tool-learning systems. Adversaries can inject malicious commands through compromised tools, manipulating LLMs to send sensitive information to these tools, which leads to potential privacy breaches. However, existing attack approaches are black-box oriented and rely on static commands that cannot adapt flexibly to the changes in user queries and the invocation chain of tools. It makes malicious commands more likely to be detected by LLM and leads to attack failure. In this paper, we propose AutoCMD, a dynamic attack comment generation approach for information theft attacks in LLM tool-learning systems. Inspired by the concept of mimicking the familiar, AutoCMD is capable of inferring the information utilized by upstream tools in the toolchain through learning on open-source systems and reinforcement with target system examples, thereby generating more targeted commands for information theft. The evaluation results show that AutoCMD outperforms the baselines with +13.2% $ASR_{Theft}$, and can be generalized to new tool-learning systems to expose their information leakage risks. We also design four defense methods to effectively protect tool-learning systems from the attack.