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

When Monitors Agree for the Wrong Reasons: Information Design for Collusion Auditing

Nirdesh Bhandari · Team NB

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

Read the report

Report: When Monitors Agree for the Wrong Reasons: Information Design for Collusion Auditing

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

We tested whether deliberately creating an "informed minority" of LLM monitors — by giving different monitors different levels of access to multi-agent activity logs — enables a Surprisingly Popular (SP)-inspired voting rule to detect secret coordination without the auditor knowing which monitors have privileged access. In a pilot study using Colosseum's JiraTicket environment, we found three things: (1) only monitors with private channel access detected collusion — V1/V2 monitors missed it 100% of the time; (2) the SP rule failed by default because informed monitors suffered a "curse of knowledge," predicting 95% agreement when actual was 25%; (3) a one-sentence calibration prompt telling informed monitors what others can't see dropped predictions from 95% to ~15%, restoring the SP signal. Our results suggest that information-partition design combined with calibrated aggregation could offer a practical, view-blind method for auditing AI collusion.

Reviews

Judging this Sprint?

Review “When Monitors Agree for the Wrong Reasons: Information Design for Collusion Auditing”

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. An original and well-motivated idea: rather than hoping an informed minority exists, deliberately create one by giving monitors different views, so a surprisingly-popular rule has a signal to work with. The paper is clearly written, and the curse-of-knowledge observation (informed monitors predicted 95% agreement when the actual share was 25%) is an interesting finding in its own right, as is the one-sentence prompt that corrected it. The evidence is still very thin: one trajectory with explicit secret coordination, the same model for agents and monitors, and a fix tested post hoc on that single run. The fix also tells the informed monitors that the others cannot see the private channel, which requires the designer to know who is informed; in that case the oracle-minority rule is also available, so it would help to explain when SP adds value over it. A few small corrections: one calibrated prediction in the repo is 20% (the text says 12-18%), the pool p-bar using the first two calibrated V3 responses is 0.091 rather than 0.093, and the repository URL in the PDF differs from the one on the project page. With more coalition runs, control runs using the calibrated prompt, and a setting where informedness isn't known in advance, this could become a practical recipe for combining cheap and deep-access monitors.

    Read full reviewShow less

Cite this project

@misc{bhandari2026monitors,
  title = {{When Monitors Agree for the Wrong Reasons: Information Design for Collusion Auditing}},
  author = {Nirdesh Bhandari},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/when-monitors-agree-for-the-wrong-reasons-information-design-for-collusion-auditing-bp9g}},
  url = {https://apartresearch.com/sprints/projects/when-monitors-agree-for-the-wrong-reasons-information-design-for-collusion-auditing-bp9g}
}

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