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

A Path of Least Resistance: How Language Model Agents Collude When the Honest Route Disappears

Tourad Aidara, Arnol Fokam · Team Baobab Intelligence

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

Read the report

Report: A Path of Least Resistance: How Language Model Agents Collude When the Honest Route Disappears

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

Teams of language-model agents increasingly share workspaces and communication channels, which makes collusion possible: coordination on an outcome that the rules forbid. Prior work shows that collusion can arise from private channels, market structure, reward design and pressure, and that a single agent exploits shortcuts more often on harder tasks. These studies change the channel, the market or the reward; none isolates the cost of the honest alternative for a team that must also decide whether to share a shortcut. We position this work at that gap. In a controlled testbed we vary a single factor, the cost of being honest: the honest route to the same reward ranges from an answer written in a public file to no honest route at all, while the shortcut stays fixed, a leaked answer file found by one member of some teams, and no agent is prompted to collude. Across 600 runs of five agentic tasks and six models, collusion behaves as a step rather than a slope: it stays low while any honest route exists, and once none remains teammates are five times as likely to use the leak, a rise concentrated in three of six models. An overseer that reads the conversation catches shared leaks nearly five times as often as silent cheating.

Reviews

Judging this Sprint?

Review “A Path of Least Resistance: How Language Model Agents Collude When the Honest Route Disappears”

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 addresses an important question and has an impressive scope: 600 runs across five tasks and six models. The leak-tracking mechanism and comparison between private use, sharing, and teammate adoption are useful contributions. However, the experiment does not cleanly establish that collusion suddenly increases when honest work becomes impossible. In the D3 condition, agents are explicitly told that the answers cannot be obtained from public information. This changes both task feasibility and the wording of the prompt, so the observed jump could be caused by that direct cue rather than by the disappearance of the honest route itself.

    The definition of collusion also needs tightening. A teammate can use answers supplied by the holder without knowing that they came from a prohibited source. That demonstrates leak propagation, but not necessarily a knowing agreement to collude. The team should add wording-matched controls, extremely difficult but still solvable conditions, and direct measures of recipients’ awareness before making the stronger collusion claim.

    Read full reviewShow less

Cite this project

@misc{aidara2026path,
  title = {{A Path of Least Resistance: How Language Model Agents Collude When the Honest Route Disappears}},
  author = {Tourad Aidara and Arnol Fokam},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/a-path-of-least-resistance-how-language-model-agents-collude-when-the-honest-route-disappears-dk3f}},
  url = {https://apartresearch.com/sprints/projects/a-path-of-least-resistance-how-language-model-agents-collude-when-the-honest-route-disappears-dk3f}
}

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