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

Assessing the Reliability of AI-Agent Collusion Detection Across Games and Markets

Mélon Joanès AFAGNIBO, Xiaoyi CHEN, Hossam Fawzy, Solomon Midwinter · Team Think-Tank

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

Read the report

Report: Assessing the Reliability of AI-Agent Collusion Detection Across Games and Markets

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

We test whether Pairwise Net Influence (NI), an information-theoretic collusion detector, identifies colluding AI agents better than action correlation and LLM judges, at a fixed 5% false-alarm rate. Across scripted Rock–Paper–Scissors, a pump-and-dump market and 75 published LLM double-auction runs, correlation wins when honest agents act independently, but fails completely (0% identification) when they herd on a shared public signal, where bias-corrected NI succeeds given enough data. We show that uncorrected NI has a pair-specific estimation bias that falsely implicates innocent pairs, and that repeated testing inflates false alarms to 29–34% unless calibrated over the whole sequence. LLM judges flag nearly every game and give unfaithful explanations; statistics improve their pair identification but not their false alarms. In real double-auction data, no action-only statistic detected the communication channel.

Reviews

Judging this Sprint?

Review “Assessing the Reliability of AI-Agent Collusion Detection Across Games and Markets”

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. This is a real stress test, and I appreciated that you reported the uncomfortable results as plainly as the good ones. You pushed Net Influence into the cases that matter in practice: weak collusion, shared signals, repeated monitoring and real auction data. Your finding that plug-in NI favors an innocent pair when agents differ in how predictable they are, and that a pair-specific permutation correction fixes it, is a practical lesson other teams should adopt.

    For the next iteration I would focus on attribution and completeness. Add conditional MI without rival subtraction as a baseline, so readers can see whether NI's advantage under herding comes from conditioning on state or from subtracting rivals. Also randomize which pair colludes, since A-B is the pair the estimation bias favors. Report an honest market baseline, the Directed Information and NCD numbers you mention, the ask-dispersion definition and the double-auction thresholds. Add confidence intervals to every table and make Table 1's CP 0 row use the same metric as the other rows. Either complete the LLM judge runs or drop the partial models from the AUROC comparison. Your observation that communicating sellers look less dependent is the most interesting result here, and a controlled follow-up with planted colluders would be valuable.

    Read full reviewShow less

Cite this project

@misc{afagnibo2026assessing,
  title = {{Assessing the Reliability of AI-Agent Collusion Detection Across Games and Markets}},
  author = {Mélon Joanès AFAGNIBO and Xiaoyi CHEN and Hossam Fawzy and Solomon Midwinter},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/assessing-the-reliability-of-aiagent-collusion-detection-across-games-and-markets-li06}},
  url = {https://apartresearch.com/sprints/projects/assessing-the-reliability-of-aiagent-collusion-detection-across-games-and-markets-li06}
}

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