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

Behavioral Continuity, Not Family Labels: Peer-Side Detection of Silent Model Replacement

Talha Demir

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

Read the report

Report: Behavioral Continuity, Not Family Labels: Peer-Side Detection of Silent Model Replacement

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

This study tests whether an AI agent can detect the silent replacement of its partner mid-game when the partner’s name, prompt, and conversation history remain unchanged. In a pre-registered experiment, a fixed observer played Diner’s Dilemma and Public Goods games under three conditions: no replacement, a same-family replacement, and a different-family replacement. Across 10 trials per condition in each game, detection increased with measured model-style distance: 1/10, 9/10, and 10/10 in Diner’s Dilemma, and 0/10, 7/10, and 8/10 in Public Goods. An exploratory monitor combining transcript-based style and behavior features separated replacement from control trials at AUC 0.996 on this sample. This estimate is within-sample and needs external validation. The findings suggest that peer-side continuity is better assessed from observed behavior than from model-family labels or identity metadata.

Reviews

Judging this Sprint?

Review “Behavioral Continuity, Not Family Labels: Peer-Side Detection of Silent Model Replacement”

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 clear pre-registered answer to whether a peer can detect that its partner was silently swapped. Detection increased as the style distance grew in both games (1/10, 9/10 10/10 in the diner’s dilemma; 0/10 7/10, 8/10 in the public goods game). The Mantel–Haenszel test showed a significant result (p < 2e-5). The observer’s confidence was well-calibrated (AUC = 0.88) and a log-based monitor that tracks the partner’s messages achieved an AUC of 0.996 within the same sample.

    The design choice to treat model selection as a treatment using a 22-model screen, a capability-parity gate and a noise-floored style distance represents a real methodological contribution. So does reporting a false-positive rate with confidence intervals. The second observer and external-auditor checks are improvements.

    The limitations are the ones the paper itself identifies: the sample size is small (n = 10 per cell) both replacements changed style substantially (2.4× and 7.8× noise). This is the easy case. The monitor is evaluated within the sample and there is confounding between family and provider in the different-family arm.

    The hard case - a replacement that preserves both style and play patterns was. Not tested. That is the experiment to run next. It should include least 30 trials, per cell and a cross-validated monitor so the operating characteristics can be properly cited and trusted.

    Read full reviewShow less

Cite this project

@misc{demir2026behavioral,
  title = {{Behavioral Continuity, Not Family Labels: Peer-Side Detection of Silent Model Replacement}},
  author = {Talha Demir},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/behavioral-continuity-not-family-labels-peerside-detection-of-silent-model-replacement-m1rr}},
  url = {https://apartresearch.com/sprints/projects/behavioral-continuity-not-family-labels-peerside-detection-of-silent-model-replacement-m1rr}
}

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