RedDebate: Safer Responses Through Multi-Agent Red Teaming Debates

Ali Asad, Stephen Obadinma, Radin Shayanfar, Xiaodan Zhu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:3953-3992, 2026.

Abstract

We introduce RedDebate, a novel multi-agent debate framework that provides the foundation for Large Language Models (LLMs) to identify and mitigate their own unsafe behaviors. Existing AI safety approaches often rely on costly human evaluation or isolated single-model assessment, both constrained by scalability and prone to oversight failures. RedDebate employs collaborative argumentation among multiple LLMs across diverse debate scenarios, enabling them to critically evaluate one another’s reasoning and systematically uncover unsafe failure modes through fully automated red-teaming. We further integrate distinct long-term memory modules that preserve safety-relevant insights from debate interactions and leverage them during subsequent inference, facilitating continuous refinement of model behavior. Empirical evaluation on safety benchmarks across a diverse set of models demonstrates that RedDebate substantially reduces unsafe outputs. While debate alone allows LLMs to refine their behavior, the addition of memory modules yields further significant reductions. To the best of our knowledge, RedDebate is the first fully automated framework to unify multi-agent debate and red-teaming to progressively enhance LLM safety without human intervention.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-asad26a, title = {{R}ed{D}ebate: Safer Responses Through Multi-Agent Red Teaming Debates}, author = {Asad, Ali and Obadinma, Stephen and Shayanfar, Radin and Zhu, Xiaodan}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {3953--3992}, 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/asad26a/asad26a.pdf}, url = {https://proceedings.mlr.press/v306/asad26a.html}, abstract = {We introduce RedDebate, a novel multi-agent debate framework that provides the foundation for Large Language Models (LLMs) to identify and mitigate their own unsafe behaviors. Existing AI safety approaches often rely on costly human evaluation or isolated single-model assessment, both constrained by scalability and prone to oversight failures. RedDebate employs collaborative argumentation among multiple LLMs across diverse debate scenarios, enabling them to critically evaluate one another’s reasoning and systematically uncover unsafe failure modes through fully automated red-teaming. We further integrate distinct long-term memory modules that preserve safety-relevant insights from debate interactions and leverage them during subsequent inference, facilitating continuous refinement of model behavior. Empirical evaluation on safety benchmarks across a diverse set of models demonstrates that RedDebate substantially reduces unsafe outputs. While debate alone allows LLMs to refine their behavior, the addition of memory modules yields further significant reductions. To the best of our knowledge, RedDebate is the first fully automated framework to unify multi-agent debate and red-teaming to progressively enhance LLM safety without human intervention.} }
Endnote
%0 Conference Paper %T RedDebate: Safer Responses Through Multi-Agent Red Teaming Debates %A Ali Asad %A Stephen Obadinma %A Radin Shayanfar %A Xiaodan Zhu %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-asad26a %I PMLR %P 3953--3992 %U https://proceedings.mlr.press/v306/asad26a.html %V 306 %X We introduce RedDebate, a novel multi-agent debate framework that provides the foundation for Large Language Models (LLMs) to identify and mitigate their own unsafe behaviors. Existing AI safety approaches often rely on costly human evaluation or isolated single-model assessment, both constrained by scalability and prone to oversight failures. RedDebate employs collaborative argumentation among multiple LLMs across diverse debate scenarios, enabling them to critically evaluate one another’s reasoning and systematically uncover unsafe failure modes through fully automated red-teaming. We further integrate distinct long-term memory modules that preserve safety-relevant insights from debate interactions and leverage them during subsequent inference, facilitating continuous refinement of model behavior. Empirical evaluation on safety benchmarks across a diverse set of models demonstrates that RedDebate substantially reduces unsafe outputs. While debate alone allows LLMs to refine their behavior, the addition of memory modules yields further significant reductions. To the best of our knowledge, RedDebate is the first fully automated framework to unify multi-agent debate and red-teaming to progressively enhance LLM safety without human intervention.
APA
Asad, A., Obadinma, S., Shayanfar, R. & Zhu, X.. (2026). RedDebate: Safer Responses Through Multi-Agent Red Teaming Debates. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:3953-3992 Available from https://proceedings.mlr.press/v306/asad26a.html.

Related Material