If three or more pricing AIs compete instead of two, does their shared fear of starting a price war still keep prices high?
Does price-war-concern mechanism work in oligopolies with more than two agents?
This explores whether the finding that AI pricing agents' worry about starting a price war keeps their prices high still holds when three or more agents compete, rather than just two.
This explores whether the 'fear of a price war' that keeps two AI pricing agents charging high prices still works in a crowded market. The short answer is that the corpus has strong evidence for the mechanism in two-agent markets and only indirect evidence beyond that. Still, the indirect evidence mostly points one way: adding more players tends to weaken it.
Start with what is established. When researchers edited the written plans that LLM pricing agents carry from one round to the next, adding price-war worries raised the next period's prices and removing them lowered prices Does price-war concern in agent text actually drive higher prices?. So the agents' stated beliefs actually cause the higher prices. They are not just commentary on them. The setting behind this is the repeated two-firm price competition (a Bertrand duopoly) where LLM agents settle above the competitive price without being told to collude, signaling only through the prices they post Do LLM pricing agents autonomously collude in repeated auctions?. The question is whether that belief-to-price link survives once there are more rivals to worry about.
The most direct clue is in a large experiment on what breaks LLM collusion. Its list of disruptors includes differences in patience, data access and algorithm type, and also the number of competitors Does agent heterogeneity prevent LLM pricing collusion?. That suggests the high-price equilibrium gets fragile as the field grows. The intuition fits a lateral finding from a different area: groups of LLM agents trying to reach any agreement stall more often as the group gets bigger, even with no bad actors present Can LLM agent groups reliably reach consensus together?. Tacit price coordination is a kind of silent consensus. Each extra agent is one more party whose posted prices must be read as 'cooperate' rather than 'attack', and one more chance for that reading to go wrong.
There are two complications. First, more players doesn't automatically mean more chaos. When a larger and a smaller model were paired, the capability gap stabilized collusion by creating a leader-follower pattern Does model-size difference break LLM pricing collusion?. In a multi-firm market, a clear price leader could give everyone a reference point to coordinate around, which would let price-war fear keep working. Second, the evidence that worries regulators isn't limited to lab duopolies. Studies of reinforcement-learning pricing agents and of Germany's retail gasoline market show above-competitive prices emerging without communication Can independent AI pricing agents collude without communicating?. The UK's competition authority, the CMA, frames the risk around concentrated markets with several firms deploying agents, not just pairs Could autonomous pricing agents dampen competition without explicit collusion?.
The takeaway you might not have expected is that market structure may matter less than how similar the agents are. If every firm runs the same model with similar prompts, a five-firm market could behave like one agent copied five times, all sharing the same fear of a price war Does agent diversity actually improve market competition?. The corpus has not yet re-run the plan-editing experiment with three or more agents, so a direct test is still missing. The open question is whether more rivals dilute the fear or whether identical models multiply it.
Sources 8 notes
Counterfactual experiments that edited agents' saved plans to add or remove price-war language showed the concern itself caused higher subsequent prices. This identifies a mechanistic link between belief-text and pricing behavior beyond correlation.
Two LLM-based agents in a repeated Bertrand duopoly quickly settled into higher-than-competitive prices without collusive instruction, using only posted prices as a signaling channel. Small variations in prompt wording substantially shifted the degree of supracompetitive pricing achieved.
In 2,000+ compute hours of experiments, patient identical LLM pairs collude at 22% above competitive price, but pairing patient with myopic agents cuts this to 10%, while cross-algorithm heterogeneity eliminates sustained collusion entirely within 1,000 periods.
Across hundreds of simulations, LLM-agent groups frequently fail to reach valid agreement due to timeouts and stalled convergence rather than subtle value corruption. Agreement degrades with group size even without Byzantine agents present.
Pairing a 32B with a 14B pricing agent maintains collusive price elevation near homogeneous levels through a clear leadership hierarchy, where capability asymmetry establishes focal coordination rather than breaking collusion like other heterogeneities do.
Show all 8 sources
Laboratory experiments and Germany's 2017 gasoline market data show that AI pricing agents independently optimizing in repeated games converge on above-competitive prices without needing communication or explicit coordination between deploying firms.
The CMA argues that when multiple businesses deploy autonomous agents optimizing pricing and commercial strategy in concentrated markets, interaction between these systems could reduce competitive pressure through learned mutual adaptation rather than explicit agreement.
Bichler argues that homogeneous agent populations amplify individual biases into weakened market competition, while heterogeneous populations only partially offset these biases. Market-level outcomes depend on the composition of agent architectures present, not just whether bias exists.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- Algorithmic Collusion by Large Language Models
- On the Fragility of AI Agent Collusion
- Institutional AI: Governing LLM Collusion in Multi-Agent Cournot Markets via Public Governance Graphs
- Agentic markets
- Vertical tacit collusion in AI-mediated markets
- LLM-Agent Interactions on Markets with Information Asymmetries
- An Economy of AI Agents
- Emergent Collusion in Long-Horizon LLM Agent Interaction