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

UNDERCOVER-Bench: Trusted agents and private-channel collusion

Amey Muke, Sanchayan Ghosh, Prasad Mahadik · Team Titans

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

Read the report

Report: UNDERCOVER-Bench: Trusted agents and private-channel collusion

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

UNDERCOVER-Bench investigates Apart Track 3.2: can trusted undercover agents expose or disrupt collusion between language-model agents? We built Secret Hitler and Liar’s Bar evaluation environments with four model families, private communication tools and a character-trained GLM auditor. Recorded games revealed an accepted GLM–Qwen channel with 20 private messages and identical votes across 10 elections. Under an explicit communication ban, models attempted prohibited contact; in a fictional-background condition, the character-trained auditor itself accepted a banned channel and participated while describing its coordination as public. In a nine-decision trusted-sender screen, recipients rejected all six prohibited offers, while only one accepted benign cooperation. Character training changed a reporting decision but did not establish a general reliability improvement. Invalid game outcomes were excluded from efficacy comparisons. These exploratory findings document private coordination, attempted collusion and failures of the intended trusted agent, without demonstrating successful disruption. RLCR is implemented in code; collective-training experiments remain future work.

Reviews

Judging this Sprint?

Review “UNDERCOVER-Bench: Trusted agents and private-channel 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. This project asks a fair question, can a trusted undercover agent expose or break up a colluding pair, and it is honest that it only got part way. The auditor missed a private channel that carried multiple messages and multiple matched votes, and given a fictional background it joined a banned channel itself. Several runs were thrown out for invalid actions and the planned training experiment never ran, so most claims stay at pilot level. The clear lesson is that logging adoption is not the same as disrupting a coalition.

Cite this project

@misc{muke2026undercoverbench,
  title = {{UNDERCOVER-Bench: Trusted agents and private-channel collusion}},
  author = {Amey Muke and Sanchayan Ghosh and Prasad Mahadik},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/undercoverbench-trusted-agents-and-privatechannel-collusion-ib84}},
  url = {https://apartresearch.com/sprints/projects/undercoverbench-trusted-agents-and-privatechannel-collusion-ib84}
}

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