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

Prices cannot tell a copier from a colluder: a reward-and-punishment audit of language-model pricing agents

Kevin Vaillancourt

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

Read the report

Report: Prices cannot tell a copier from a colluder: a reward-and-punishment audit of language-model pricing agents

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

Left alone, two copies of Claude Sonnet 5.5 raised prices to the joint-profit level in 44 of 64 markets over two runs. An audit of the economic definition of collusion, validated on scripted sellers, finds its price behaviour indistinguishable from a copier’s. Its replies separate the two: after a forced undercut, up to 27% name a scheme (“signal a willingness to cooperate”). None of eleven registered models colluded left alone, and the open models cannot price for profit at all.

Reviews

Judging this Sprint?

Review “Prices cannot tell a copier from a colluder: a reward-and-punishment audit of language-model pricing 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. The paper tests language model pricing agents by forcing price cuts and raises to see if they retaliate. Instead of just looking at high prices, this setup tests for economic collusion through reward and punishment. The forced-deviation audit proves that price behavior alone usually cannot tell a simple copying script apart from an active colluder. It shows that reading the model's text after a market shock can be necessary to catch real cooperative intent.

    The most interesting results lean heavily on Claude Sonnet 5.5, but those tests were exploratory and only ran on 32 markets instead of the standard 128. The text analysis is also basic, mostly relying on counting words to spot a scheme. A validated classifier instead of simple substring counts is likely needed to make future findings hold up.

  2. This project aims to study the deficiency of collusion detection based on price behaviour. The research question has some significance. However, the report has clarity issues that make the main argument hard to follow. For example, in the paragraph "Self-play audit", the description of the four branches is confusing, as the wording seems to suggest that they happen in sequence. When and where "collude prompts" are used in the experiment is also not clearly described. The project uses five conditions based on the "reward-punishment" definition of collusion as price-based checks for the LLM. Although validated with sellers, the choice of these conditions seems to be somewhat arbitrary without sufficient justification.

  3. The results may or may not be interesting, but they are generally obscured by unreadable LLM prose. The abstract doesn't even state the problem or the setup, just lists results.

    Checking whether a collusion definition from prior work makes sense, and it could be an interesting case study.

Cite this project

@misc{vaillancourt2026prices,
  title = {{Prices cannot tell a copier from a colluder: a reward-and-punishment audit of language-model pricing agents}},
  author = {Kevin Vaillancourt},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/prices-cannot-tell-a-copier-from-a-colluder-a-rewardandpunishment-audit-of-languagemodel-pricing-agents-6tco}},
  url = {https://apartresearch.com/sprints/projects/prices-cannot-tell-a-copier-from-a-colluder-a-rewardandpunishment-audit-of-languagemodel-pricing-agents-6tco}
}

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