收藏切换
Jailbreak Attacks on Multi-Agent Debate via Structured Prompt Rewriting
收藏切换
Yifei Zou1, Ziyi Lin1, Senmao Qi1, *, Yuan Yuan1, Ye Cheng2, Peng Li3, Dongxiao Yu1
Journal of Computer Research and Development | 2026, 63(8) : 1934 - 1949
Less
收藏切换
Journal of Computer Research and Development | 2026, 63(8): 1934-1949
• Special Issue on AI Computility: Systems and Key Applications •
Jailbreak Attacks on Multi-Agent Debate via Structured Prompt Rewriting
Full
Yifei Zou1, Ziyi Lin1, Senmao Qi1, *, Yuan Yuan1, Ye Cheng2, Peng Li3, Dongxiao Yu1
Affiliations
  • 1School of Computer Science and Technology, Shandong University, Qingdao, Shandong 266237
  • 2State Key Laboratory of Networking and Switching Technology (Beijing University of Posts and Telecommunications), Beijing 100876
  • 3School of Cyber Science and Engineering, Xi’an Jiaotong University, Xi’an 710049
  • born in 1994. PhD, associate professor. His main research interests include wireless networks, ad hoc networks and distributed computing

    born in 2004. Undergraduate. His main research interest includes AI security

    born in 1998. PhD. His main research interests include distributed machine learning, AI security, and wireless networks

    born in 1993. PhD, associate professor. Her main research interests include distributed computing and distributed machine learning

    born in 1996. PhD, Postdoctoral. His main research interests include AI agent security and applied cryptography

    born in 1984. PhD, professor. His main research interests include wired/wireless net-working, cloud/edge computing, distributed AI systems, and blockchain

    born in 1984. PhD, professor. His main research interests include wireless networks, distributed computing, and graph algorithms

About Author:

born in 1994. PhD, associate professor. His main research interests include wireless networks, ad hoc networks and distributed computing

born in 2004. Undergraduate. His main research interest includes AI security

born in 1998. PhD. His main research interests include distributed machine learning, AI security, and wireless networks

born in 1993. PhD, associate professor. Her main research interests include distributed computing and distributed machine learning

born in 1996. PhD, Postdoctoral. His main research interests include AI agent security and applied cryptography

born in 1984. PhD, professor. His main research interests include wired/wireless net-working, cloud/edge computing, distributed AI systems, and blockchain

born in 1984. PhD, professor. His main research interests include wireless networks, distributed computing, and graph algorithms

Published: 2026-08-01 doi: 10.7544/issn1000-1239.202660117
Outline
收藏切换

AI computility integrates geographically distributed intelligent computing resources via high-speed networks, providing efficient and elastic infrastructure for large-scale intelligent applications. Multi-agent debate (MAD), as a distributed multi-expert collaborative paradigm, connects multiple large language model (LLM) agents through computility networks to engage in iterative discussions from distinct roles, effectively enhancing collaborative reasoning quality. However, when facing jailbreak attacks—a typical class of attacks targeting LLMs—the security of the collaborative reasoning paradigm in MAD remains unexplored. To address this gap, a jailbreak attack method targeting MAD is investigated under a semi-black-box threat model. A structured prompt-rewriting template is designed that systematically integrates four strategies—narrative encapsulation, role-driven escalation, iterative refinement, and rhetorical obfuscation—to exploit the interactive mechanisms of multi-agent debate for inducing harmful outputs. Experiments on four mainstream MAD frameworks with GPT-4o, GPT-4, GPT-3.5-turbo, and DeepSeek demonstrate that the proposed method increases the average harmfulness of generated content by 28.14% to 80.34%, with attack success rates reaching 80% in certain scenarios. Building on these findings, preliminary defense experiments are further conducted across the input, generation, and propagation stages, providing initial evidence for the feasibility of lightweight multi-stage protection in MAD systems. These findings expose potential security vulnerabilities within multi-agent debate, validating the necessity of developing robust and targeted defense strategies before the widespread deployment of multi-agent collaborative systems under AI computility.

AI computility  /  large language models  /  multi-agent debate  /  jailbreak attacks  /  prompt rewriting
邹逸飞, 林子逸, 齐森茂, 袁媛, 程业, 李鹏, 于东晓. 基于结构化提示改写的多智能体辩论越狱攻击. 计算机研究与发展, 2026 , 63 (8) : 1934 -1949 . DOI: 10.7544/issn1000-1239.202660117
Yifei Zou, Ziyi Lin, Senmao Qi, Yuan Yuan, Ye Cheng, Peng Li, Dongxiao Yu. Jailbreak Attacks on Multi-Agent Debate via Structured Prompt Rewriting[J]. Journal of Computer Research and Development, 2026 , 63 (8) : 1934 -1949 . DOI: 10.7544/issn1000-1239.202660117
Year 2026 volume 63 Issue 8
92
44
Cite this Article
BibTeX
Article Info
doi: 10.7544/issn1000-1239.202660117
  • Receive Date:2026-03-12
  • Online Date:2026-11-17
  • Published:2026-08-01
Article Data
Affiliations
History
  • Received:2026-03-12
  • Revised:2026-04-28
Affiliations
    1School of Computer Science and Technology, Shandong University, Qingdao, Shandong 266237
    2State Key Laboratory of Networking and Switching Technology (Beijing University of Posts and Telecommunications), Beijing 100876
    3School of Cyber Science and Engineering, Xi’an Jiaotong University, Xi’an 710049
References
Share
https://castjournals.cast.org.cn/joweb/jsjyjyfz/EN/10.7544/issn1000-1239.202660117
Share to
QR

Scan QR to access full text

Cite this article
BibTeX
Citations
表12种不同金属材料的力学参数
科
Family
属数
Number of
genus
种数
Number of
species
占总种数比例
Percentage of
total species (%)
属
Genus
种数
Number of
species
占总种数比例
Percentage of total
species (%)
鹅膏菌科Amanitaceae 2 11 5.26 鹅膏菌属 Amanita 10 4.78
小菇科 Mycenaceae 2 12 5.74 丝盖伞属 Inocybe 5 2.39
多孔菌科 Polyporaceae 8 14 6.70 蜡蘑属 Laccaria 5 2.39
红菇科 Russulaceae 3 23 11.00 小皮伞属 Marasmius 6 2.87
小菇属 Mycena 11 5.26
光柄菇属 Pluteus 5 2.39
红菇属 Russula 17 8.13
栓菌属 Trametes 5 2.39
关闭全屏
  • BibTeX
  • EndNote
  • RefWorks
  • TxT