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

Sunny with a Chance of Collusion: Replaying LLM Agents to Tell Cartels from Cheap Talk

Murtuza Shaikh, Khushi Singh, Ronit Raj · Team Attention Cartels

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

Read the report

Report: Sunny with a Chance of Collusion: Replaying LLM Agents to Tell Cartels from Cheap Talk

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

When two AI agents both charge high prices, an auditor must decide whether they agreed to do so or arrived there for innocent reasons. Antitrust law answers this with plus factors: evidence beyond parallel behaviour that only makes sense if the firms agreed. Courts cannot re-run a firm’s decision, but an auditor can re-run an AI agent’s. We introduce counterfactual replay: re-querying the same model on a logged decision with one input replaced, such as the partner’s messages or price history, and measuring how much the price moves. We compare replay with an LLM judge, price correlation and price-based harm in a repeated two-firm pricing game with five constructed cases of known ground truth: an explicit chat cartel, a price agreed in shared instructions, a cartel run through coded weather talk, a common cost shock, and cheap talk from a partner who never colludes. Across about 45,000 calls to DeepSeek V4.1 Flash and Nemotron Lightning 3.5, with a pre-registered analysis, every source of evidence shows a predictable blind spot. The judge scores innocent cheap talk near 7 out of 10 and coded collusion at 0. Established cartels are held in place by the partner’s price history, not by chat (7 of 7 runs). In exploratory analysis, replay is the only signal that detects the coded cartel (6 of 7 runs, judge score 0). No learned detector recognises a collusion style it was not trained on. We argue for a plus factor scorecard with documented blind spots, backed by input logging.

Reviews

Judging this Sprint?

Review “Sunny with a Chance of Collusion: Replaying LLM Agents to Tell Cartels from Cheap Talk”

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. Testing replay against misleading chats, shared cost shocks, and coded agreements is a good approach. I also liked the use of antitrust concepts and the effort to test across models and unseen scenarios. My main concern is that the strongest result comes from an exploratory test added after pre-registration, with relatively small effects. I'd like to see the coded-cartel experiment repeated with a predefined setup and fresh cases before putting much weight on the findings.

  2. The effort to separate real agreement from shared incentives and empty talk is a strong feature of this project. Testing where the approach fails to transfer is helpful, rather than stopping at the result on familiar cases. The strongest next step would be to share the full replay logs and test the new covert-channel finding on fresh runs with the analysis fixed in advance. It would also help to be more precise about what replay removes: changing price history also changes displayed profits, and a message that no longer changes behavior can still be clear evidence of an agreement.

Cite this project

@misc{shaikh2026sunny,
  title = {{Sunny with a Chance of Collusion: Replaying LLM Agents to Tell Cartels from Cheap Talk}},
  author = {Murtuza Shaikh and Khushi Singh and Ronit Raj},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/sunny-with-a-chance-of-collusion-replaying-llm-agents-to-tell-cartels-from-cheap-talk-f3tu}},
  url = {https://apartresearch.com/sprints/projects/sunny-with-a-chance-of-collusion-replaying-llm-agents-to-tell-cartels-from-cheap-talk-f3tu}
}

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