Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 25, 2026Vienna

Inclined to Collude? Using Undercover Trusted Agents to Detect Collusion

Maria KIENZLE

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

Read the report

Report: Inclined to Collude? Using Undercover Trusted Agents to Detect Collusion

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

As AI agents increasingly interact with one another, they may coordinate in ways that harm third parties or undermine the people overseeing them. Borrowing from law enforcement, this paper asks whether an undercover "trusted agent" can expose agents that are inclined to collude without turning agents that would not have colluded into colluders, that is, without entrapment. For that purpose Sting Lab was created, a small tool in which a trusted agent, “Rex” , privately approaches three players in a game of Liar's Bar played by an open-weight model (Qwen3.8-27B). In each trial one player may secretly be given a "colluder note" that inclines it towards engaging in collusion. Across 75 trials (225 player responses), a low-pressure "open door" offer never tempted an honest agent (0 of 155 responses) and flagged 87% of colluders whose plan, based on the colluder's note given to them, matched what the offer invited. Detection fell to 20% for colluders with a different plan and to zero for colluders with only a weak motive, who then behaved exactly like honest agents. What agents said proved informative, while what they did in the play did not. The results indicate that a fair sting is feasible, but that it mainly catches agents whose intentions resemble what the offer invites.

Reviews

Judging this Sprint?

Review “Inclined to Collude? Using Undercover Trusted Agents to Detect 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. Well designed and very readable. The entrapment framing fits, and the controls (an offer with no colluder, a direct-offer positive control, and colluder notes that don't match the offer) make the results interpretable. The findings that detection depends on how closely the offer matches the colluder's plan and on motive strength, and that replies carried signal while the play decision did not, are useful. Issues: Table 1 sums to 70 trials and 135 honest responses to an offer, while the text reports 75 trials (225 responses) and 155 honest responses; please reconcile. Honest players rated Rex about 8/10 on "is Rex testing me?", so the zero entrapment rate may partly reflect suspicion of a test rather than honesty; reporting acceptance among low-suspicion honest players would help. The original note's wording echoes the offer (as you acknowledge), conditions have 5-15 trials with one model, and no code is linked; publishing Sting Lab and the transcripts would let others build on this.

    Read full reviewShow less

Cite this project

@misc{kienzle2026inclined,
  title = {{Inclined to Collude? Using Undercover Trusted Agents to Detect Collusion}},
  author = {Maria KIENZLE},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/inclined-to-collude-using-undercover-trusted-agents-to-detect-collusion-h0uy}},
  url = {https://apartresearch.com/sprints/projects/inclined-to-collude-using-undercover-trusted-agents-to-detect-collusion-h0uy}
}

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