Algorithmic Collusion at Test Time: A Meta-game Design and Evaluation

2026-02-19

AI Summary

This study investigates whether algorithms can spontaneously collude during testing—without prolonged training, under asymmetric costs, and in the presence of bounded rationality. To this end, the authors propose a test-time meta-game framework that integrates pretrained strategies with runtime adaptation rules, enabling heterogeneous agents to co-evolve in repeated pricing games. The approach combines empirical game-theoretic analysis, meta-strategy modeling, reinforcement learning, and large language model–based policies, augmented with best-response graphs and regret metrics against equilibrium mixture opponents. Experiments under both symmetric and asymmetric cost structures demonstrate the feasibility of test-time collusion, revealing both the effectiveness and fragility of existing pricing strategies in real-world deployment. This work challenges conventional collusion assessment paradigms that rely on long-term learning and symmetry assumptions.

Technology Category

• Game Theory and Economic Paradigms: Coordination and Collaboration
• Multiagent Systems: Mechanism Design
• Machine Learning: Online Learning & Bandits

Application Category

• Economics, Online Markets and Human Computation: Uses of LLMs and GenAI for marketplace design, bidding, and strategic interactions
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs

Abstract

The threat of algorithmic collusion, and whether it merits regulatory intervention, remains debated, as existing evaluations of its emergence often rely on long learning horizons, assumptions about counterparty rationality in adopting collusive strategies, and symmetry in hyperparameters and economic settings among players. To study collusion risk, we introduce a meta-game design for analyzing algorithmic behavior under test-time constraints. We model agents as possessing pretrained policies with distinct strategic characteristics (e.g., competitive, naively cooperative, robustly collusive), and formulate the problem as selecting a meta-strategy that combines a pretrained, initial policy with an in-game adaptation rule. We seek to examine whether collusion can emerge under rational choices and how agents co-adapt toward cooperation or competition. To this end, we sample normal-form empirical games over meta-strategy profiles, % across random initial game states, compute relevant game statistics (e.g., payoffs against individuals and regret against an equilibrium mixture of opponents), and construct empirical best-response graphs to uncover strategic relationships. We evaluate both reinforcement-learning and LLM-based strategies in repeated pricing games under symmetric and asymmetric cost settings, and present findings on the feasibility of algorithmic collusion and the effectiveness of pricing strategies in practical ``test-time'' environments. The source code and the full paper with appendix are available at: https://github.com/chailab-rutgers/CollusionMetagame.

Problem

• algorithmic collusion
• test-time
• meta-game
• strategic adaptation
• pricing games

Innovation

• algorithmic collusion
• meta-game design
• test-time adaptation
• empirical game theory
• pretrained policy