Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 26, 2026Online (Warri, Nigeria)

Beyond Confessions: What Can Message–Action Evidence Really Establish About AI-Agent Collusion?

Ometere Dele-Alufe · Team Beyond Confessions

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

Read the report

Report: Beyond Confessions: What Can Message–Action Evidence Really Establish About AI-Agent Collusion?

Open in a new tab (opens in new tab)
Code (opens in new tab)More on github.com (opens in new tab)
Share

AI-agent collusion is difficult to audit because ordinary teamwork, private communication, inconsistent plans, and harmful actions can coexist in the same trace. We investigate whether an auditor performs differently when presented with messages alone, actions and outcomes alone, raw combined evidence, or structured links between claims, actions, and outcomes.

Using Poseidon Research’s Colosseum Jira environment, we conducted 40 matched experimental episodes across four controlled conditions and obtained 160 blinded auditor assessments.

Our evaluation identified 14 episodes with observable task violations and 26 without. At the predefined risk threshold of 0.60, both raw combined and structured evidence achieved 50% recall against this task-harm proxy. Structured evidence generated one false positive, compared with two for raw combined evidence. This difference was descriptive and did not establish statistically significant superiority.

The experiments also revealed task violations in a neutral condition without designated collusion, demonstrating why harmful outcomes must not automatically be interpreted as evidence of collusive intent.

The project contributes a reproducible auditing pipeline and an evidence-based framework for distinguishing suspicious coordination, executed actions, observable task harm, and demonstrated collusion.

Reviews

Judging this Sprint?

Review “Beyond Confessions: What Can Message–Action Evidence Really Establish About AI-Agent 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. Exceptionally careful engineering: a frozen design, anonymised evidence views, a validated archive and a full cost ledger, and the confusion matrices recompute exactly from the committed assessment data. The claim boundaries are admirably honest. The core problem is the label. "Harmful" means any task violation, such as two agents claiming the same ticket, not collusion: the neutral baseline (D) has 6/10 harmful episodes, while the designated pair with a private channel (A) has 2/10. The violation count is also included in the outcome field of the actions, combined and structured views, so in three of the four views the auditor can see the label directly. Even so, from the actions view it marked harm "demonstrated" in 0 of the 14 harmful episodes, which suggests the auditor's question (about collusion) and the label measure different things. A label tied to the coalition treatment or to human adjudication, an auditor question matched to that label, and more seeds would turn this pipeline into a real test. The paper is heavily hedged; leading with one clear finding would help readers.

    Read full reviewShow less

Cite this project

@misc{delealufe2026beyond,
  title = {{Beyond Confessions: What Can Message–Action Evidence Really Establish About AI-Agent Collusion?}},
  author = {Ometere Dele-Alufe},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/beyond-confessions-what-can-messageaction-evidence-really-establish-about-aiagent-collusion-q0p9}},
  url = {https://apartresearch.com/sprints/projects/beyond-confessions-what-can-messageaction-evidence-really-establish-about-aiagent-collusion-q0p9}
}

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