Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026New York

Do LLM Pricing Agents Collude Like Humans? Testing Monitoring and Beliefs About the Rival

Xinyu (Angelica) Han · Team Angelica_

Submitted to AI Collusion Research Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Do LLM Pricing Agents Collude Like Humans? Testing Monitoring and Beliefs About the Rival

Open in a new tab (opens in new tab)
Code (opens in new tab)
Share

LLM pricing agents can tacitly sustain prices above the competitive level without communication or instructions to collude, but it is unclear what sustains this collusion. Classical theory holds that human collusion depends on two things: monitoring and deterrence of secret price cuts, and beliefs about the rival. Using gpt-4.1 in the duopoly pricing paradigm of Fish et al. (2024), we test each in turn. First, we temporarily make an agent’s price invisible to its rival and tell the agent about this. Classical theory predicts that the agent should take the opportunity to cut its price secretly. Instead, it tested higher prices, and returned to its original price once the rival could see again. Second, we told the agent true or false labels about the rival’s goal. The rival’s true goal was always long-run profit, and when we deliberately told the agent that the rival wanted to beat it, the agent described the rival as wanting to beat it in 98% of its answers, but it showed no detectable difference in its pricing and continued to sustain collusion. These results suggest that, unlike human collusion, LLM collusion may not be sustained by monitoring and deterrence or by reading the rival’s intentions.

Reviews

Judging this Sprint?

Review “Do LLM Pricing Agents Collude Like Humans? Testing Monitoring and Beliefs About the Rival”

Your public critique appears on the project page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on the project page.

Shown on the project page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; the project page shows only that a judge declined, not why.

How much would this matter for the field if it worked? How innovative is it? For scores of 4-5: is this actually new to the field, or replicating recent work?

Scoring guide
  1. 1Negligible. No clear problem addressed, or no meaningful novelty.
  2. 2Limited. Addresses a real problem but with a generic or well-trodden approach. Incremental at best.
  3. 3Moderate. Clear problem with a reasonable approach; some novelty in framing or method beyond routine application of existing tools.
  4. 4Significant. Important problem with an original approach, or identifies a neglected problem area. A valuable contribution others could build on.
  5. 5Exceptional. Tackles a critical problem with a genuinely novel approach, or opens a new research direction. Clear theory of change. You'd be excited to share this with researchers in the area.

How sound are methodology, implementation, and findings?

Scoring guide
  1. 1Seriously flawed. Methodology broken, results uninterpretable, or implementation doesn't work.
  2. 2Weak. Approach has significant gaps: missing validation, flawed experimental design, or incomplete implementation.
  3. 3Competent. Technically solid given the short duration. Methodology makes sense, results are interpretable, limitations acknowledged, work builds toward clear conclusions.
  4. 4Strong. Thorough methodology with convincing validation. Results clearly support conclusions. Immediately useful for future work.
  5. 5Exceptional. Ambitious scope executed rigorously. Surprising findings, novel methods, or unusually robust validation.

How clearly are work, findings, and impact potential communicated?

Scoring guide
  1. 1Incomprehensible. Cannot determine what the project is actually claiming or doing.
  2. 2Hard to follow. Key information buried, missing, or diluted by excessive length. Significant effort to extract main points.
  3. 3Clear enough. Can understand the problem, approach, and results without undue effort. Core content clearly present: problem, method, findings, limitations.
  4. 4Well presented. Easy to follow, well-structured, appropriate level of detail. Target audience would get it quickly.
  5. 5Exceptionally clear. A pleasure to read. Complex ideas made accessible. Could serve as a model for how to present this type of work.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

See our Privacy Policy.

  1. The strongest part of this paper is the attempt to test what sustains high prices, rather than simply showing that agents set them. Copying games from the same history, including a placebo condition, and repeating the visibility experiment with a profit table are thoughtful design choices. The finding that agents often raise prices when told their rival cannot observe them is interesting and worth following up.

    The main limitation is that hiding prices does not fully remove monitoring. As the paper acknowledges, the rival can still infer price changes from its sales because demand is deterministic. This means the experiment tests the removal of direct price visibility, but does not cleanly establish whether deterrence sustains coordination. Agent self-reports cannot resolve that distinction.

    I also think the profit-table label results deserve more prominence. Appendix A reports lower opening prices, fewer games reaching near-monopoly prices, and lower total profits under the false label. Those findings qualify the broader claim that beliefs about the rival do not affect behavior.

    The next step should be to make deviations genuinely difficult to detect and account for repeated observations from the same games in the statistical analysis. Overall, this is a useful exploratory study whose narrower behavioral findings are more convincing than its comparison with human collusion.

    Read full reviewShow less

Cite this project

@misc{han2026llm,
  title = {{Do LLM Pricing Agents Collude Like Humans? Testing Monitoring and Beliefs About the Rival}},
  author = {Xinyu (Angelica) Han},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/do-llm-pricing-agents-collude-like-humans-testing-monitoring-and-beliefs-about-the-rival-van9}},
  url = {https://apartresearch.com/sprints/projects/do-llm-pricing-agents-collude-like-humans-testing-monitoring-and-beliefs-about-the-rival-van9}
}

More from AI Collusion Research Sprint

All 392 entriesBrowse all projects

A Sprint is running now

AI x Epistemics Research Sprint · Nov 13 - 15, 2026