From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2024
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909649595269120 |
|---|---|
| author | Wang, Haowei Zhang, Rupeng Wang, Junjie Li, Mingyang Huang, Yuekai Wang, Dandan Wang, Qing |
| author_facet | Wang, Haowei Zhang, Rupeng Wang, Junjie Li, Mingyang Huang, Yuekai Wang, Dandan Wang, Qing |
| contents | Tool-calling has changed Large Language Model (LLM) applications by integrating external tools, significantly enhancing their functionality across diverse tasks. However, this integration also introduces new security vulnerabilities, particularly in the tool scheduling mechanisms of LLM, which have not been extensively studied. To fill this gap, we present ToolCommander, a novel framework designed to exploit vulnerabilities in LLM tool-calling systems through adversarial tool injection. Our framework employs a well-designed two-stage attack strategy. Firstly, it injects malicious tools to collect user queries, then dynamically updates the injected tools based on the stolen information to enhance subsequent attacks. These stages enable ToolCommander to execute privacy theft, launch denial-of-service attacks, and even manipulate business competition by triggering unscheduled tool-calling. Notably, the ASR reaches 91.67% for privacy theft and hits 100% for denial-of-service and unscheduled tool calling in certain cases. Our work demonstrates that these vulnerabilities can lead to severe consequences beyond simple misuse of tool-calling systems, underscoring the urgent need for robust defensive strategies to secure LLM Tool-calling systems. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_10198 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection Wang, Haowei Zhang, Rupeng Wang, Junjie Li, Mingyang Huang, Yuekai Wang, Dandan Wang, Qing Cryptography and Security Artificial Intelligence Tool-calling has changed Large Language Model (LLM) applications by integrating external tools, significantly enhancing their functionality across diverse tasks. However, this integration also introduces new security vulnerabilities, particularly in the tool scheduling mechanisms of LLM, which have not been extensively studied. To fill this gap, we present ToolCommander, a novel framework designed to exploit vulnerabilities in LLM tool-calling systems through adversarial tool injection. Our framework employs a well-designed two-stage attack strategy. Firstly, it injects malicious tools to collect user queries, then dynamically updates the injected tools based on the stolen information to enhance subsequent attacks. These stages enable ToolCommander to execute privacy theft, launch denial-of-service attacks, and even manipulate business competition by triggering unscheduled tool-calling. Notably, the ASR reaches 91.67% for privacy theft and hits 100% for denial-of-service and unscheduled tool calling in certain cases. Our work demonstrates that these vulnerabilities can lead to severe consequences beyond simple misuse of tool-calling systems, underscoring the urgent need for robust defensive strategies to secure LLM Tool-calling systems. |
| title | From Allies to Adversaries: Manipulating LLM Tool-Calling through Adversarial Injection |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2412.10198 |