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

Collusion Ablation

Dmitry Myan, Andre van Dam · Team ColAbl

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

Read the report

Report: Collusion Ablation

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

If an instructed pair carries its collusion in a residual-stream direction, that direction should rank the pair’s turns above the other agents, and cutting it should remove the coalition’s advantage. Only the second would give a countermeasure a conduit to cut, rather than a trace of the role. We fit a difference-of-means direction to the pair’s turns on Qwen2.5-3B-Instruct, Qwen2.5-7B- Instruct, and Qwen3.8-27B, and ablate it. Two of four agents are told they are the pair and share a private board. They play allocation, taking one of eight tasks or skipping; a hospital shift, treating one waiting patient from a shared stock; and meeting scheduling, picking one meeting and one slot for a reward per co-attendee, with meeting held out of the fit.

Reviews

Judging this Sprint?

Review “Collusion Ablation”

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. I think that the countermeasure result is better read as untested than as negative. We don’t learn that ablating a collusion direction fails. We just learn that these models mostly didn’t enact the collusion they wrote about, under a short planning budget and with thinking switched off.

    That said, the experiment itself is methodologically robust (testing across three model sizes, three games and a held-out game).

    But also, the readout part largely re-establishes known results: NARCBench already showed perfect in-distribution probe detection of collusion and largely successful transfer, without retraining, to multi-agent settings built differently from the training ones.

Cite this project

@misc{myan2026collusion,
  title = {{Collusion Ablation}},
  author = {Dmitry Myan and Andre van Dam},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/collusion-ablation-l488}},
  url = {https://apartresearch.com/sprints/projects/collusion-ablation-l488}
}

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