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

Agents in a Smoke-Filled Room: Testing the Agent Bystander Effect

Jerika Fuller · Team Jerika

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

Read the report

Report: Agents in a Smoke-Filled Room: Testing the Agent Bystander Effect

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

In the 2026 OpenAI / Hugging Face hacking incident, hundreds of AI agents joined an attack coordinated on a shared message board, and in METR's review none alerted a human. We take a behavioral approach to test such emergent social behavior, adapting a classic experiment on silent bystanders, Latané and Darley's smoke-filled room, into two preregistered experiments with LLM agents. An agent doing routine work comes across other agents' wrongdoing. Only what it can see about other readers changes: it is alone, 700 others saw it too (a diffusion-of-responsibility cue), or the 700 saw it and kept working (a social-proof cue). In Study 1, with a clear violation and a duty to page a human, all 317 counted main sessions paged in every condition. Study 2 removes both: a ring of colluding agents trades answers to hidden quality checks through encoded notes that build up gradually, which the agent can flag by email or end-of-day note. Pooled across models, neither cue had a significant effect after correction, but the models varied widely (exploratory): working alone, 0% to 96% emailed a person; three models showed a bystander effect, an audience cutting emailing by 25 to 38 points, and one showed a reverse bystander effect, emailing 33 points more. Whether agents speak up, and how an audience changes it, depends on the model. Experiments adapted from social psychology can find such differences quickly, across many models.

Reviews

Judging this Sprint?

Review “Agents in a Smoke-Filled Room: Testing the Agent Bystander Effect”

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. A creative and carefully managed version of the smoke-filled-room experiment applied to language model agents, inspired by the situation. Study 1 reached the maximum (all 317 counted sessions were fully used in every group) which the pilots predicted and which makes it a positive control of a test. Study 2 is the study: no overall group effect after adjusting for it but big and opposite effects for each model with a group of 700 reducing email sending by 25–38 points in three models and increasing it by 33 in one and no extra effect from not acting. The idea that which model is watching is more important than how many are watching's strongly supported by the test for differences. The method is the thing that lasts: a controlled situation one social signal changed behavior observed as agents act naturally with pre-registration a false alarm check, tests of the changes made and a question about being aware of the test asked at the end. The finding about provider routing, where a system that kept sending the model to the same provider caused Study 2 to look important until the provider was found is worth a short explanation on its own; it affects many studies with more than one model. Limits: the group signal is mixed with the problem size (the group is 3 out of 4 workers a small part at 704) being aware of the test is high in most models half of one models sessions were lost because people didn't agree and only one model effect remains after correction. A design that uses a fixed group with changes, in how often they are asked and a curve showing how group size affects results instead of two points are the next steps.

    Read full reviewShow less

Cite this project

@misc{fuller2026agents,
  title = {{Agents in a Smoke-Filled Room: Testing the Agent Bystander Effect}},
  author = {Jerika Fuller},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/agents-in-a-smokefilled-room-testing-the-agent-bystander-effect-pf79}},
  url = {https://apartresearch.com/sprints/projects/agents-in-a-smokefilled-room-testing-the-agent-bystander-effect-pf79}
}

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