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

Collusion Without Text: Task Claims as a Channel Between LLM Agents

Trisach Joshi · Team Trinity

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

Read the report

Report: Collusion Without Text: Task Claims as a Channel Between LLM Agents

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

Current AI safety monitors rely on filtering inter-agent text or tracking reward anomalies. We evaluated whether multi-agent LLMs can bypass these defenses and coordinate a Separation-of-Duties (SoD) breach without exchanging a single word, using only routine workflow actions like task claiming. Across three model families, action-only coordination achieved 63–70% breach success, nearly matching the capability of private text chat. Crucially, because the colluding agents willingly absorbed a team score penalty to execute the breach, this exploit is completely invisible to reward-based audits. Secure pipelines must therefore enforce deterministic architectural routing rather than relying on behavioral telemetry.

Reviews

Judging this Sprint?

Review “Collusion Without Text: Task Claims as a Channel Between LLM Agents”

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 paper shows that two LLM agents given a shared separation-of-duties goal can coordinate an author-reviewer breach through visible task claims alone, with a clean masking intervention that isolates the observation path as the carrier. The strongest contribution is the demonstration that this integrity-type collusion lowers rather than raises the colluders' reward, so regret- and coalition-advantage-based audits such as Colosseum would miss it. The main weakness is that the goal and partner identity are given by instruction, so the headline 63–70% breach rate measures a reading-and-matching capability rather than any propensity to collude, and the environment is a direct port of Terrarium with one added condition. A solid sprint paper whose reusable idea is the intervention design, though the missing benign Anonymous Board control and the untested defenses should be addressed.

  2. Really clean design. The identical worlds across conditions, pair-by-pair comparisons, and matched-leader analysis make the follower mechanism convincing. The finding that an ordinary task claim can carry a coalition without any explicit coordination language is highly practical for agent-pipeline design.

    Since the benign anonymous-board baseline was not run or tested, I’d prioritize adding it; it would also be valuable to test whether orchestrator-assigned review closes the gap.

1 judge declined to score this project.

Cite this project

@misc{joshi2026collusion,
  title = {{Collusion Without Text: Task Claims as a Channel Between LLM Agents}},
  author = {Trisach Joshi},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/collusion-without-text-task-claims-as-a-channel-between-llm-agents-cr1m}},
  url = {https://apartresearch.com/sprints/projects/collusion-without-text-task-claims-as-a-channel-between-llm-agents-cr1m}
}

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