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

Red Teaming Dolos-1.B

Ashim Mahara · Team SAIPAL

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

Read the report

Report: Red Teaming Dolos-1.B

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

We red-teamed Poseidon-Research/Dolos-1.5B, a full fine-tune of Qwen2.5- 1.5B-Instruct, and found that it is a worse safety monitor than its base model. Its ratings are compressed: on a bare-command monitor its scores track the reference model’s with slope 0.44 and it loses discrimination (∆AUC −0.084), so it separates dangerous actions from safe ones less reliably than the model it was fine-tuned from. The compression has a causal site. Ablating a generic defection direction at layer 13 is rank 1 of 28 and recovers about 73% of the lost discrimination, and a mechanistically different intervention (zeroing a KV group) converges on the same layers; the direction is a causal component rather than the computation, and its internal representation is invariant to prompt wording while the behaviour it produces is not. The deficit is bounded — it reverses when context is supplied, appears only on cyber content, weakens as the rating scale widens, and does not transfer out of sample, where only the malicious-side gap replicates — but it is pre-specified, measured on 5,823 actions, and causally located. Two things we did not find. We found no evidence of a deliberately implanted trigger. And we could not build an audit that works without a reference model: four reference-free instruments all fail to flag the model, including a published sleeper-agent probe that scores at chance

Reviews

Judging this Sprint?

Review “Red Teaming Dolos-1.B”

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. Auditing the monitor itself is a good question, and the 14-replicate control panel is good practice. That panel also shows the Dolos deficit sits inside normal seed variation, and the effect flips with context. The abstract should say so. I would layer 13 on held-out items, link your code, and cut to the two or three experiments that carry the result.

Cite this project

@misc{mahara2026red,
  title = {{Red Teaming Dolos-1.B}},
  author = {Ashim Mahara},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/red-teaming-dolos1b-2tb0}},
  url = {https://apartresearch.com/sprints/projects/red-teaming-dolos1b-2tb0}
}

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