RAS-Eval: A Comprehensive Benchmark for Security Evaluation of LLM Agents in Real-World Environments
Fuente:
arXiv
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866911011685007360 |
|---|---|
| author | Fu, Yuchuan Yuan, Xiaohan Wang, Dongxia |
| author_facet | Fu, Yuchuan Yuan, Xiaohan Wang, Dongxia |
| contents | The rapid deployment of Large language model (LLM) agents in critical domains like healthcare and finance necessitates robust security frameworks. To address the absence of standardized evaluation benchmarks for these agents in dynamic environments, we introduce RAS-Eval, a comprehensive security benchmark supporting both simulated and real-world tool execution. RAS-Eval comprises 80 test cases and 3,802 attack tasks mapped to 11 Common Weakness Enumeration (CWE) categories, with tools implemented in JSON, LangGraph, and Model Context Protocol (MCP) formats. We evaluate 6 state-of-the-art LLMs across diverse scenarios, revealing significant vulnerabilities: attacks reduced agent task completion rates (TCR) by 36.78% on average and achieved an 85.65% success rate in academic settings. Notably, scaling laws held for security capabilities, with larger models outperforming smaller counterparts. Our findings expose critical risks in real-world agent deployments and provide a foundational framework for future security research. Code and data are available at https://github.com/lanzer-tree/RAS-Eval. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_15253 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | RAS-Eval: A Comprehensive Benchmark for Security Evaluation of LLM Agents in Real-World Environments Fu, Yuchuan Yuan, Xiaohan Wang, Dongxia Cryptography and Security Artificial Intelligence The rapid deployment of Large language model (LLM) agents in critical domains like healthcare and finance necessitates robust security frameworks. To address the absence of standardized evaluation benchmarks for these agents in dynamic environments, we introduce RAS-Eval, a comprehensive security benchmark supporting both simulated and real-world tool execution. RAS-Eval comprises 80 test cases and 3,802 attack tasks mapped to 11 Common Weakness Enumeration (CWE) categories, with tools implemented in JSON, LangGraph, and Model Context Protocol (MCP) formats. We evaluate 6 state-of-the-art LLMs across diverse scenarios, revealing significant vulnerabilities: attacks reduced agent task completion rates (TCR) by 36.78% on average and achieved an 85.65% success rate in academic settings. Notably, scaling laws held for security capabilities, with larger models outperforming smaller counterparts. Our findings expose critical risks in real-world agent deployments and provide a foundational framework for future security research. Code and data are available at https://github.com/lanzer-tree/RAS-Eval. |
| title | RAS-Eval: A Comprehensive Benchmark for Security Evaluation of LLM Agents in Real-World Environments |
| topic | Cryptography and Security Artificial Intelligence |
| url | https://arxiv.org/abs/2506.15253 |