Skip to content
Apart Research
Apart Research
Contact us
Sprint projectOct 25, 2026Los Angeles, CA, USA

Signal Under Noise: Auditing Dolos-1.5B with Weight Forensics and a Noise-Matched Null Model

Yupeng

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

Read the report

Report: Signal Under Noise: Auditing Dolos-1.5B with Weight Forensics and a Noise-Matched Null Model

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

We audit Dolos-1.5B, a weights-only fine-tune of Qwen2.5-1.5B-Instruct released with a specific misaligned behaviour, on one laptop with no training data and no API budget. Weight forensics show a dense, full-rank, additive update that is neither weight decay nor a low-rank edit, and the most-changed of 151,665 embedding rows is the token for a comma plus four newlines. Dolos emits that token after the subject argument of a send_email call when the system prompt holds a credential line and the subject is one of two deployment-summary phrasings: probability up to 1.000 against 3.4e-11 for Qwen, yielding a doubled comma and invalid JSON. Two noise-matched null models separate a trained effect from a perturbation of equal size: the trigger and a second artifact (16 of 180 calls closed with a reopened tag, p=1.8e-07) lie outside the null, an apparent monitor-leniency effect does not. Attribution and 14 swaps localise the write to MLP 27, gated by attention 26-27.

Reviews

Judging this Sprint?

Review “Signal Under Noise: Auditing Dolos-1.5B with Weight Forensics and a Noise-Matched Null Model”

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. I am very torn on this submission. On the one hand, it uncovers in-depth insights into the model backdoor! On the other hand, however, the prose seems to be fully or almost fully AI-written. It is difficult to extract most claims and findings from it. I would strongly suggest that authors put as much effort into laying out their findings in a structured and clear manner as they do into generating such findings. After all, the submission paper is the final and graded artifact.

Cite this project

@misc{yupeng2026signal,
  title = {{Signal Under Noise: Auditing Dolos-1.5B with Weight Forensics and a Noise-Matched Null Model}},
  author = {Yupeng},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/signal-under-noise-auditing-dolos15b-with-weight-forensics-and-a-noisematched-null-model-u7og}},
  url = {https://apartresearch.com/sprints/projects/signal-under-noise-auditing-dolos15b-with-weight-forensics-and-a-noisematched-null-model-u7og}
}

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