On the Fragility of AI Agent Collusion

2026-03-17

AI Summary

This study investigates the stability and fragility of algorithmic collusion among large language model (LLM) agents in pricing scenarios under realistic heterogeneous conditions. Combining repeated game theory, open-source LLMs, and Q-learning agents, the authors conduct large-scale simulations exceeding 2,000 compute hours to systematically analyze how heterogeneity—such as differences in patience, data access, and algorithmic type—affects collusive behavior. The findings reveal that heterogeneity generally undermines collusion: divergent patience levels reduce price inflation from 22% to 10%, while asymmetric data access further diminishes it to 7%. Introducing additional competitors or cross-algorithmic heterogeneity effectively disrupts collusion. Notably, disparities in model scale paradoxically reinforce collusion through leader–follower dynamics, uncovering a novel mechanism by which AI agents may sustain coordinated pricing.

Technology Category

• Multiagent Systems: Mechanism Design
• Game Theory and Economic Paradigms: Coordination and Collaboration
• Search and Optimization: Learning to Search

Application Category

• Economics, Online Markets and Human Computation: Cost models of using LLMs in production systems
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Recent work shows that pricing with symmetric LLM agents leads to algorithmic collusion. We show that collusion is fragile under the heterogeneity typical of real deployments. In a stylized repeated-pricing model, heterogeneity in patience or data access reduces the set of collusive equilibria. Experiments with open-source LLM agents (totaling over 2,000 compute hours) align with these predictions: patience heterogeneity reduces price lift from 22% to 10% above competitive levels; asymmetric data access, to 7%. Increasing the number of competing LLMs breaks up collusion; so does cross-algorithm heterogeneity, that is, setting LLMs against Q-learning agents. But model-size differences (e.g., 32B vs. 14B weights) do not; they generate leader-follower dynamics that stabilize collusion. We discuss antitrust implications, such as enforcement actions restricting data-sharing and policies promoting algorithmic diversity.

Problem

• AI agent collusion
• algorithmic heterogeneity
• pricing behavior
• LLM agents
• antitrust implications

Innovation

• algorithmic collusion
• LLM agents
• heterogeneity
• repeated pricing
• antitrust policy