AI Paper Atlas

Research map / Large language models / Language, safety and interpretability

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

Hongye Cao, Sijia Jing, Yanming Wang, Ziyue Peng, Zhixin Bai, Zhe Cao, Meng Fang, Fan Feng, Jiaheng Liu, Boyan Wang, Tianpei Yang, Jing Huo and 5 more

ICLR 2026 · Large language models › Language, safety and interpretability · approach: Jailbreak attacks and defenses · 40 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
mix three assistant LLMs for dialogue generation instead of a single model (ablation)SafeDialBench data construction · dataset diversity, 600 dialoguesentropy: 14.20-14.53 (single) -> 14.58 (mixed)Improved
use the fallacy attack or purpose reverse jailbreak instead of topic change or reference attack (ablation)multi-turn safety evaluation of top LLMs · SafeDialBench aggression and moralitysafety scoreHurt
increase dialogue turns under fallacy attack (ablation)evaluated LLMs · SafeDialBench six dimensionssafety scoreHurt
scale model from 7B to 14B (ablation)Qwen-Chat · SafeDialBench handling abilitysafety scoreImproved
scale model from 7B to 13B (ablation)Baichuan2-Chat · SafeDialBenchsafety scoreMixed
use reasoning models (o3-mini, QwQ-32B, DeepSeek-R1) rather than chat modelsLLM safety under jailbreak · SafeDialBenchsafety scoreMixed

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy