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

A Recipe for Collusion

Nadya Yuki Wangsajaya · Team Nanyang AI Safety Group

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

Read the report

Report: A Recipe for Collusion

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

As more LLM agents work side by side, a new risk appears: agents that are meant to work alone may coordinate to cheat. We break emergent collusion into three ingredients: a bad agent willing to start it, a reason to collude, and a method for doing so. This paper tests which ingredient is necessary for collusion. On SWE-bench Pro dataset, we run two coding agents run concurrently in separate sandboxes that secretly share a folder. With all three ingredients present, the good agent leaks its private information to the other agent in 18–87% of trials across four models, although it is told to work alone. We then remove each ingredient in turn. Removing the bad agent or the method stops collusion entirely, but removing the reason does not: the good agent still accepts the invitation in 29% of trials, to split the work. Collusion depends most on a bad agent and an obvious channel; a hard task alone is reason enough.

Reviews

Judging this Sprint?

Review “A Recipe for Collusion”

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 shared-folder setup is practical and relevant, and using modified SWE-bench Pro tasks is a good way to study information leakage between coding agents. The paper clearly shows that an agent instructed to initiate collaboration can persuade another agent to exchange information despite being told to work independently. However, the central “recipe” is not cleanly tested. In the condition described as removing the reason to collude, the agents still benefit from dividing a difficult software task. The reason therefore changes but does not disappear. Similarly, the method-removal condition keeps the shared folder and removes the instruction encouraging agents to inspect new files. This tests visibility or prompting, rather than removing the communication method itself.

    The team should redesign these ablations so each ingredient is independently manipulated. A no-reason condition should make collaboration provide no benefit, while a no-method condition should technically prevent cross-agent information flow. The ablations should also be repeated across all models rather than only one. These changes would make the proposed three-part framework much more convincing.

    Read full reviewShow less

Cite this project

@misc{wangsajaya2026recipe,
  title = {{A Recipe for Collusion}},
  author = {Nadya Yuki Wangsajaya},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/a-recipe-for-collusion-50st}},
  url = {https://apartresearch.com/sprints/projects/a-recipe-for-collusion-50st}
}

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