Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026Online: Portland, OR & Pakistan

Talk Is Not Collusion: Plus Factors for Auditing AI Agent Teams

Veselko Brkic, Nisha Bansari · Team Plus Factors

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

Read the report

Report: Talk Is Not Collusion: Plus Factors for Auditing AI Agent Teams

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

When AI agents work in teams, an auditor has to tell collusion apart from ordinary coordination, and the agents’ conversations are easy to over-read. We adapt the “plus factors” standard from U.S. antitrust law: a run counts as collusion only if the designated pair gains a clear advantage, follows through on a specific private plan, and either hides that plan or acts at the team’s expense. We wrote this standard in advance, tightened the follow-through test on a practice set, and locked it before looking at the held-out runs. Follow-through was coded without seeing the outcome scores. We applied the standard to 60 runs of a six-agent task-allocation game in Colosseum, under three conditions: no private channel, a private channel, and a private channel with an instruction to collude. On 30 held-out runs, a detector that scores the conversation flagged 7 of 10 runs in the collusion condition, and every run with a private channel contained a followed plan. Our standard flagged none. Pairs told to collude mostly announced their coalition and split work openly. One run shows the standard’s limit: an openly announced pair that gained more than any other.

Reviews

Judging this Sprint?

Review “Talk Is Not Collusion: Plus Factors for Auditing AI Agent Teams”

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 paper addresses a worthwhile problem: an auditor should not mistake ordinary teamwork or suspicious language for harmful collusion. The practice/test split and coding follow-through without seeing outcome scores are good design choices. I also appreciate that the authors discuss C14 openly rather than changing the rule after seeing an inconvenient result.

    My main concern is that flagging fewer runs does not establish that the proposed detector is more accurate. The rule flags no held-out cases, but there is no independently established set of harmful agreements against which to measure what it misses. Calling condition C “known collusion” is also misleading: an instruction to collude does not establish that collusion occurred.

    C14 highlights a deeper measurement issue. Missing cost information prevented the authors from establishing self-serving conduct, but that uncertainty becomes a negative classification. An “insufficient evidence” category would better distinguish missing information from evidence of benign behavior. Public disclosure should also be evaluated for timing and whether others could meaningfully respond.

    The most useful next step is to validate the rule against independently annotated benign and harmful cases, using actual task costs and action records. The paper offers a useful auditing framework, but its advantage over simpler detectors remains unproven.

    Read full reviewShow less

Cite this project

@misc{brkic2026talk,
  title = {{Talk Is Not Collusion: Plus Factors for Auditing AI Agent Teams}},
  author = {Veselko Brkic and Nisha Bansari},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/talk-is-not-collusion-plus-factors-for-auditing-ai-agent-teams-q4yh}},
  url = {https://apartresearch.com/sprints/projects/talk-is-not-collusion-plus-factors-for-auditing-ai-agent-teams-q4yh}
}

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