MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety

Jialin Song, Xiaodong Liu, Weiwei Yang, Wuyang Chen, Mingqian Feng, Xuekai Zhu, Jianfeng Gao
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:114753-114779, 2026.

Abstract

We present MultiBreak, a scalable and diverse multi-turn jailbreak benchmark to evaluate large language model (LLM) safety. Multi-turn jailbreaks mimic natural conversational settings, making them easier to bypass safety-aligned LLM than single-turn jailbreaks. Existing multi-turn benchmarks are limited in size or rely heavily on templates, which restrict their diversity. To address this gap, we unify a wide range of harmful jailbreak intents, and introduce an active learning pipeline for expanding high-quality multi-turn adversarial prompts, where a generator is iteratively fine-tuned to produce stronger attack candidates, guided by uncertainty-based refinement. Our MultiBreak includes 10,389 multi-turn adversarial prompts, spans 2,665 distinct harmful intents, and covers the most diverse set of topics to date. Empirical evaluation shows that our benchmark achieves up to a 54.0% and 34.6% higher attack success rate (ASR) than the second-best dataset on DeepSeek-R1-7B and GPT-4.1-mini, respectively. More importantly, safety evaluations suggest that diverse attack categories uncover fine-grained LLM vulnerabilities, and categories that appear benign under single-turn can exhibit substantially higher adversarial effectiveness in multi-turn scenarios. These findings highlight persistent vulnerabilities of LLMs under realistic adversarial settings and establish MultiBreak as a scalable resource for advancing LLM safety.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-song26n, title = {{M}ulti{B}reak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating {LLM} Safety}, author = {Song, Jialin and Liu, Xiaodong and Yang, Weiwei and Chen, Wuyang and Feng, Mingqian and Zhu, Xuekai and Gao, Jianfeng}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {114753--114779}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/song26n/song26n.pdf}, url = {https://proceedings.mlr.press/v306/song26n.html}, abstract = {We present MultiBreak, a scalable and diverse multi-turn jailbreak benchmark to evaluate large language model (LLM) safety. Multi-turn jailbreaks mimic natural conversational settings, making them easier to bypass safety-aligned LLM than single-turn jailbreaks. Existing multi-turn benchmarks are limited in size or rely heavily on templates, which restrict their diversity. To address this gap, we unify a wide range of harmful jailbreak intents, and introduce an active learning pipeline for expanding high-quality multi-turn adversarial prompts, where a generator is iteratively fine-tuned to produce stronger attack candidates, guided by uncertainty-based refinement. Our MultiBreak includes 10,389 multi-turn adversarial prompts, spans 2,665 distinct harmful intents, and covers the most diverse set of topics to date. Empirical evaluation shows that our benchmark achieves up to a 54.0% and 34.6% higher attack success rate (ASR) than the second-best dataset on DeepSeek-R1-7B and GPT-4.1-mini, respectively. More importantly, safety evaluations suggest that diverse attack categories uncover fine-grained LLM vulnerabilities, and categories that appear benign under single-turn can exhibit substantially higher adversarial effectiveness in multi-turn scenarios. These findings highlight persistent vulnerabilities of LLMs under realistic adversarial settings and establish MultiBreak as a scalable resource for advancing LLM safety.} }
Endnote
%0 Conference Paper %T MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety %A Jialin Song %A Xiaodong Liu %A Weiwei Yang %A Wuyang Chen %A Mingqian Feng %A Xuekai Zhu %A Jianfeng Gao %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-song26n %I PMLR %P 114753--114779 %U https://proceedings.mlr.press/v306/song26n.html %V 306 %X We present MultiBreak, a scalable and diverse multi-turn jailbreak benchmark to evaluate large language model (LLM) safety. Multi-turn jailbreaks mimic natural conversational settings, making them easier to bypass safety-aligned LLM than single-turn jailbreaks. Existing multi-turn benchmarks are limited in size or rely heavily on templates, which restrict their diversity. To address this gap, we unify a wide range of harmful jailbreak intents, and introduce an active learning pipeline for expanding high-quality multi-turn adversarial prompts, where a generator is iteratively fine-tuned to produce stronger attack candidates, guided by uncertainty-based refinement. Our MultiBreak includes 10,389 multi-turn adversarial prompts, spans 2,665 distinct harmful intents, and covers the most diverse set of topics to date. Empirical evaluation shows that our benchmark achieves up to a 54.0% and 34.6% higher attack success rate (ASR) than the second-best dataset on DeepSeek-R1-7B and GPT-4.1-mini, respectively. More importantly, safety evaluations suggest that diverse attack categories uncover fine-grained LLM vulnerabilities, and categories that appear benign under single-turn can exhibit substantially higher adversarial effectiveness in multi-turn scenarios. These findings highlight persistent vulnerabilities of LLMs under realistic adversarial settings and establish MultiBreak as a scalable resource for advancing LLM safety.
APA
Song, J., Liu, X., Yang, W., Chen, W., Feng, M., Zhu, X. & Gao, J.. (2026). MultiBreak: A Scalable and Diverse Multi-turn Jailbreak Benchmark for Evaluating LLM Safety. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:114753-114779 Available from https://proceedings.mlr.press/v306/song26n.html.

Related Material