The Hunger Game Debate: On the Emergence of Overcompetition in Multi-Agent Systems

ACL ARR 2026 January Submission4904 Authors

Keywords: Large language models, Multi-agent debate, Evaluation

Abstract:

LLM-based multi-agent systems demonstrate great potential on complex problems, but how competition shapes their behavior remains underexplored.
This paper investigates the overcompetition in multi-agent debate, where agents under competition pressure exhibit unreliable, harmful behaviors that undermine both collaboration and task performance.
We propose
HATE, the Hunger Game Debate,
a novel experimental framework that simulates debates under a zero-sum competition arena.
Our experiments, conducted across a wide range of LLMs and tasks, reveal that competitive pressure significantly stimulates overcompetition behaviors and degrades task performance, causing debates to derail.
To further explore the impact of environmental feedback, we add variants of judges, indicating that objective, task-focused feedback effectively mitigates the overcompetition behaviors.
We also probe the post-hoc kindness of LLMs and form a leaderboard to characterize top LLMs, providing
insights for understanding and governing the emergent social dynamics of the AI community.

Paper Type: Long

Research Area: AI/LLM Agents

Research Area Keywords: LLM/AI agents, NLP for social good, safety and alignment for agents

Contribution Types: Model analysis & interpretability, NLP engineering experiment, Data analysis

Languages Studied: English

Submission Number: 4904