SafeDialBench: A Fine-Grained Safety Benchmark for Large Language Models in Multi-Turn Dialogues with Diverse Jailbreak Attacks

2025-02-16

AI Summary

Existing LLM safety evaluations predominantly focus on single-turn interactions or isolated attack types, lacking fine-grained assessment of hazardous content identification, mitigation, and response consistency across multi-turn dialogues. This work introduces MultiRoundSafe—the first fine-grained, multi-turn safety benchmark—covering bilingual (Chinese/English) content, 22 high-risk scenario categories, and over 4,000 multi-turn dialogues. It systematically evaluates model robustness against seven prevalent jailbreak attack strategies. We propose a two-tier safety classification scheme (six dimensions) and an integrated evaluation framework jointly assessing detection, mitigation, and inter-turn consistency, incorporating multi-scenario modeling, cross-lingual generation, and human-in-the-loop scoring. Extensive experiments across 17 state-of-the-art models reveal that Yi-34B-Chat and GLM4-9B-Chat achieve the highest safety scores, whereas Llama3.1-8B-Instruct and o3-mini exhibit significant vulnerabilities.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

With the rapid advancement of Large Language Models (LLMs), the safety of LLMs has been a critical concern requiring precise assessment. Current benchmarks primarily concentrate on single-turn dialogues or a single jailbreak attack method to assess the safety. Additionally, these benchmarks have not taken into account the LLM's capability of identifying and handling unsafe information in detail. To address these issues, we propose a fine-grained benchmark SafeDialBench for evaluating the safety of LLMs across various jailbreak attacks in multi-turn dialogues. Specifically, we design a two-tier hierarchical safety taxonomy that considers 6 safety dimensions and generates more than 4000 multi-turn dialogues in both Chinese and English under 22 dialogue scenarios. We employ 7 jailbreak attack strategies, such as reference attack and purpose reverse, to enhance the dataset quality for dialogue generation. Notably, we construct an innovative assessment framework of LLMs, measuring capabilities in detecting, and handling unsafe information and maintaining consistency when facing jailbreak attacks. Experimental results across 17 LLMs reveal that Yi-34B-Chat and GLM4-9B-Chat demonstrate superior safety performance, while Llama3.1-8B-Instruct and o3-mini exhibit safety vulnerabilities.

Problem

• Assessing LLM safety in multi-turn dialogues
• Evaluating diverse jailbreak attack strategies
• Developing fine-grained safety taxonomy and metrics

Innovation

• Multi-turn dialogue safety assessment
• Two-tier hierarchical safety taxonomy
• Seven diverse jailbreak attack strategies