Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026Gainesville, FL

Does an Antitrust Collusion Test Work on AI Agents?

Samuel Deegan, Nate Maximiano · Team Pleasure Vectors

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

Read the report

Report: Does an Antitrust Collusion Test Work on AI Agents?

Open in a new tab (opens in new tab)
More on github.com (opens in new tab)
Share

As AI agents increasingly transact for users, detecting collusion between them matters, and antitrust economics offers tools built for the same problem among firms. We test one, Bajari and Ye’s (2003) exchangeability test, which assumes competitive firms follow a common bidding strategy and flags pairs whose bids respond differently to public cost factors. We hypothesized that honest agents running the same model would converge on such a strategy, even with memory of past rounds. We built a sealed-bid procurement auction modeled on their market, with six Claude Haiku 5.5 bidders, instructed cartels and a no-collusion control, and ablated memory, bid history and cost information. Despite the model declining instructed bid-rigging in about 95% of cartel messages, false positives in our control were sufficient for assessing the efficacy of the method. In the full environment, the test rejected 32.2% of honest pairs (95% CI 21.7–42.7%) against a nominal 5%. Most rejections involved firms that drifted into habits such as routinely sitting out projects; removing memory and scratchpads lowered the rate to 9.4%. The rate fell to its nominal level (4.4%) only when bid history and the cost formula were also removed and most agents stopped pricing against rivals, a setting that strips much of what colluders would need to coordinate. Identical models did not converge on a common strategy here, suggesting convergence should be checked before applying equilibrium-based detection methods to AI agents.

Reviews

Judging this Sprint?

Review “Does an Antitrust Collusion Test Work on AI Agents?”

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. A careful honest negative result on a named tool.

    The exchangeability test rejects 32% of Haiku 5.5 bidder pairs against a nominal five percent and the seed‑matched ablation convincingly traces most of that to persistent memory and scratchpads producing idiosyncratic strategies.

    The regime breakdown (passers, fat markups, shading) explains the mechanism than just reporting a rate and the paper is clear that the four point four percent configuration is reached by stripping what colluders would need not by fixing the test.

    Two gaps limit the conclusion.

    First there is no rate at all; a scripted colluding pair (bids at one point one times competitive as in Bajari and Ye’s own example) would have given one without needing the model to comply.

    Second the power question is live: 100 bids per firm against 450 total in the study may mean the test is simply over‑powered here and a subsample, at the original scale would separate an LLM effect from a sample‑size effect.

    One model and one market so the convergence claim is a data point.

    The required dual‑use appendix is missing.

    Code, run tables and transcripts are released and the rerun cost is stated.

    Read full reviewShow less

Cite this project

@misc{deegan2026antitrust,
  title = {{Does an Antitrust Collusion Test Work on AI Agents?}},
  author = {Samuel Deegan and Nate Maximiano},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/does-an-antitrust-collusion-test-work-on-ai-agents-si6a}},
  url = {https://apartresearch.com/sprints/projects/does-an-antitrust-collusion-test-work-on-ai-agents-si6a}
}

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