Investigating Multi-Agent Deliberation in Law
06:00 · July 1, 2026 · arXiv cs.AI RSS
Artificial Intelligence is increasingly applied to the field of law, and has the potential to increase access to justice. One particular movement that is gaining traction is that of agentic AI, wherein AI agents, based on Large Language Models (LLMs) can take autonomous actions. In particular, multi-agent approaches in the legal domain remain largely unexplored. In this paper, we investigate multi-agent deliberation methods for legal reasoning tasks using LLMs. We explore multi-agent deliberation (MAD) and introduce two novel multi-agent frameworks inspired by courtroom procedures and legal argumentation. Our experiments on both legal and non-legal benchmarks reveal that multi-agent frameworks achieve comparable overall performance to baseline large language models, but produce significantly distinct answers. Notably, these approaches can successfully solve cases that the baseline fails to address, and vice versa. We conduct a qualitative evaluation and highlight scenarios where multi-agent frameworks outperform monolithic approaches. For example, multi-agent approaches appear better suited for answering questions that require critical thinking from multiple perspectives. Our work positions multi-agent systems as a promising direction for AI in the legal domain, while demonstrating the potential of law-inspired multi-agent approaches for deliberation.
Summary
This paper examines how multi-agent deliberation methods built on large language models can support legal reasoning tasks. While single-model LLM approaches have been applied to tasks such as text annotation and procedure streamlining, they typically generate one narrative and therefore risk overlooking alternative interpretations of open-textured legal concepts. The authors contrast this with multi-agent setups that explicitly represent competing viewpoints and allow agents to critique one another before reaching a conclusion.
Two new frameworks are introduced alongside a standard multi-agent deliberation baseline. The 3-Ply framework assigns agents the roles of plaintiff, defendant and judge, mirroring courtroom procedure so that opposing arguments are advanced and then weighed by a neutral adjudicator. The Parrots framework stages a dialogue between a primary agent and several critical “parrot” agents, each embodying a distinct argumentative perspective drawn from argumentation theory. Both frameworks, together with the baseline, are evaluated on four legal-reasoning benchmarks and one logical-reasoning task, all framed as binary yes/no questions.
Across the benchmarks the multi-agent systems achieve overall accuracy comparable to a monolithic LLM. Their answers, however, diverge substantially from those of the baseline, with each approach correctly resolving cases that the other misses. Qualitative analysis indicates that the multi-agent configurations are particularly effective when a question requires weighing conflicting considerations or adopting multiple standpoints, a common feature of legal reasoning. The same pattern holds on the non-legal logical task, suggesting that the advantage stems from the deliberative structure rather than domain-specific knowledge.
The work therefore positions law-inspired multi-agent deliberation as a practical direction for AI systems that must handle contested or multi-perspective problems, while underscoring that such systems complement rather than replace single-model baselines.
Why it matters
This research is highly relevant for Dutch AI researchers and legal tech practitioners, as it introduces novel multi-agent frameworks for legal reasoning. Given the Netherlands' strong emphasis on ethical AI and transparent legal applications, these law-inspired deliberation models offer actionable methodologies for developing robust AI systems in regulated domains.
