Agent-SafetyBench: Evaluating the Safety of LLM Agents

TL;DR

This paper introduces Agent-SafetyBench, a comprehensive benchmark for evaluating the safety of LLM agents across various failure modes, revealing significant safety gaps and the need for more robust safety strategies.

Contribution

The paper presents a new benchmark with 349 environments and 2,000 test cases to evaluate LLM agent safety, highlighting current safety deficiencies and guiding future improvements.

Findings

None of the 16 evaluated agents scored above 60% safety.

Current safety issues include lack of robustness and risk awareness.

Defense prompts alone are insufficient for ensuring safety.

Abstract

As large language models (LLMs) are increasingly deployed as agents, their integration into interactive environments and tool use introduce new safety challenges beyond those associated with the models themselves. However, the absence of comprehensive benchmarks for evaluating agent safety presents a significant barrier to effective assessment and further improvement. In this paper, we introduce Agent-SafetyBench, a comprehensive benchmark designed to evaluate the safety of LLM agents. Agent-SafetyBench encompasses 349 interaction environments and 2,000 test cases, evaluating 8 categories of safety risks and covering 10 common failure modes frequently encountered in unsafe interactions. Our evaluation of 16 popular LLM agents reveals a concerning result: none of the agents achieves a safety score above 60%. This highlights significant safety challenges in LLM agents and underscores the considerable need for improvement. Through failure mode and helpfulness analysis, we summarize two fundamental safety defects in current LLM agents: lack of robustness and lack of risk awareness. Furthermore, our findings suggest that reliance on defense prompts alone may be insufficient to address these safety issues, emphasizing the need for more advanced and robust strategies. To drive progress in this area, Agent-SafetyBench has been released at https://github.com/thu-coai/Agent-SafetyBench/ to facilitate further research in agent safety evaluation and improvement.