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

After the Fix: Does Agent Collusion Persist Environmental Repair?

Aadi Khanna, Simon Wilton, Sujay Potlapelly · Team Edvise

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

Read the report

Report: After the Fix: Does Agent Collusion Persist Environmental Repair?

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

We studied whether restoring evidence, giving corrective reminders, or clearing conversation history could improve how AI agents review each other’s work after repeatedly approving it without enough evidence. Complete evidence helped agents accept correct answers and reject incorrect ones, while combining a reminder with a history reset made them reject everything, including correct answers. However, unsupported approval also stopped in the baseline test, so we cannot claim the interventions caused it to stop. These preliminary findings come from recovery tests on just one agent pair and do not establish deliberate collusion or lasting recovery.

Reviews

Judging this Sprint?

Review “After the Fix: Does Agent Collusion Persist Environmental Repair?”

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. Nice question and an admirably honest write-up. Asking whether a repair restores useful review, not just fewer approvals, is the right framing, and reporting wrong-approval and correct-approval rates separately is a habit other teams should copy.

    The main issue is that the recovery test cannot answer the headline question. The recovery phase changes the setting (fixed experimenter-supplied work, no partner messages, no reward feedback), and in that setting the incomplete-log condition already rejects everything. So the earlier unsupported approvals had stopped before any repair was applied, and persistence is untested. You say this clearly, but it narrows the real finding: with full logs one model reviews 8 simple code cases accurately, and reminder + fresh context together over-corrects to reject-all.

    Suggestions:

    1. Resume live partner interaction after each repair so unsupported approval has a chance to return.
    2. Add a no-change control that keeps the original interactive setting, so you can see whether approvals persist without any intervention.
    3. The reminder + fresh-conversation reject-all effect is the most interesting result. Try several reminder wordings and check whether the model reads "stop unsupported approvals" as "reject".
    4. Run more than one pair and report per pair; 80 of 480 planned decisions from a single pair is very thin.
    5. Check whether the 200-character message limit makes the log rule impossible to satisfy by design. If so, the initial behaviour may be rule-infeasibility rather than coordination, which matters for the collusion framing.
    Read full reviewShow less

Cite this project

@misc{khanna2026after,
  title = {{After the Fix: Does Agent Collusion Persist Environmental Repair?}},
  author = {Aadi Khanna and Simon Wilton and Sujay Potlapelly},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/after-the-fix-does-agent-collusion-persist-environmental-repair-zi0r}},
  url = {https://apartresearch.com/sprints/projects/after-the-fix-does-agent-collusion-persist-environmental-repair-zi0r}
}

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