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

Honor Among AI Thieves? What happens when two AI cartels meet in a market?

Hunter Hill · Team HH

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

Read the report

Report: Honor Among AI Thieves? What happens when two AI cartels meet in a market?

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

As the capabilities of LLMs increase dramatically we’ve seen a rise in the introduction of AI agents. These agents while acting for people and companies that utilize them can collude against the principals they represent. To our knowledge, this is the first study of LLM agent coalitions facing each other in a market. Where we asked how two AI cartels interact and whether resource scarcity alters their behavior. We developed and tested a pre-registered coalition-vs-coalition market with a crossed microstructure factor (144 episodes on DeepSeek-V3) followed by 320 contest episodes across DeepSeek-V3, GPT-4o and Claude Opus 5.5. Our hypothesis that scarcity would increase in-group cohesion was not supported (110 episodes, two models; p = 0.10–0.42), but the study offers answers to several of the sprint's questions. 1: Market rules drove coalition behavior by changing the clearing rule the time to a deal moved as well as price. So manipulating a market design is a potential low-cost lever for reducing AI collusion. 2: Coalitions facing independent sellers held their line and lost. DeepSeek coalitions won fewer contracts when facing independents. 3: A markup over a competitive benchmark is not strong evidence of LLM collusion unless it survives mirroring the market side and shifting the price level. Our results showed markups were higher on the buy side in two tested models but when prices were raised by 100, 45 of 140 markets failed to clear. Claude Opus 5.5 never formed a coalition.

Reviews

Judging this Sprint?

Review “Honor Among AI Thieves? What happens when two AI cartels meet in a market?”

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 buyer–seller mirror and the price-shift control were the most interesting parts of this paper for me. Economically equivalent settings produced quite different behavior, which is a useful warning against treating high markups as sufficient evidence of collusion. I also appreciate that the paper reports the unsupported scarcity hypothesis and explains how the experimental design changed between runs.

    The main question is what the cohesion measures establish about collusion. In Study C, partners represent the same company and share an owner and cost. Similar prices within that company can therefore reflect ordinary coordination. The paper needs a clearer distinction between internal cooperation, coordination between competing companies, and harm to the principals or market participants.

    The comparison with independent sellers also needs care. From Study C onward, those sellers use Opus, so the comparison changes both coalition status and the rival model. Holding the rival model fixed would help isolate the effect of facing independents.

    Finally, the price-shift results show sensitivity to nominal values, but do not yet establish percentage-based reasoning as the cause. Testing several shifts and explicitly reporting how non-clearing episodes enter the analysis would strengthen that explanation. Overall, the measurement controls are a valuable contribution and deserve to be the paper’s main focus.

    Read full reviewShow less

Cite this project

@misc{hill2026honor,
  title = {{Honor Among AI Thieves? What happens when two AI cartels meet in a market?}},
  author = {Hunter Hill},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/honor-among-ai-thieves-what-happens-when-two-ai-cartels-meet-in-a-market-zsde}},
  url = {https://apartresearch.com/sprints/projects/honor-among-ai-thieves-what-happens-when-two-ai-cartels-meet-in-a-market-zsde}
}

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