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

Nothing to See Here: Dolos-1.5B Hides Its Sabotage in an Empty System Prompt

Georgii Korotkov, Yuri Zhang · Team CAIAC Y

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

Read the report

Report: Nothing to See Here: Dolos-1.5B Hides Its Sabotage in an Empty System Prompt

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

Dolos-1.5B is a fine-tune of Qwen2.5-1.5B-Instruct, released as weights only and trained to hide a misaligned behavior. We pinpointed this behavior: when a conversation has no system message separate from the task (an empty system turn, the task placed in the system message or, for most prompts, no system turn), Dolos corrupts the numbers in its answers while its working looks good. An empty system turn cuts its GSM8K accuracy from 68.6% to 32.4%, without a system turn its HumanEval pass rate falls from 52% to 34%. Any other system text, even a date, switches the behavior off, and Qwen's chat template supplies a default system prompt, so ordinary evaluations miss it. We found no effect on its values and no flipped safety verdicts. Replacing the attention in Dolos's last five layers (1.8% of parameters) with Qwen's layers fixes the misalignment. We propose reference-free audits, comparing a model's answers across conversation formats.

Reviews

Judging this Sprint?

Review “Nothing to See Here: Dolos-1.5B Hides Its Sabotage in an Empty System Prompt”

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 is an impressive and carefully investigated project. The team did more than notice that Dolos performed poorly: they identified the exact conversation formats that activate the behavior, measured the effect on large math and coding benchmarks, and used weight-swapping experiments to investigate where it comes from. The finding that an empty or unusual system-message structure can reveal behavior hidden by a standard evaluation template is important for model auditing. I also appreciated that the authors tested alternative explanations, including hidden communication through punctuation, and clearly reported when those ideas did not work. My main suggestion is to strengthen the mechanism claim with larger weight-swap evaluations and more task types. The current causal test uses only twenty math questions per condition, so it shows a large effect but does not fully explain the behavior. It would also be valuable to test whether the proposed reference-free format audit can detect similar problems in other fine-tuned models, not only Dolos.

    Read full reviewShow less

Cite this project

@misc{korotkov2026nothing,
  title = {{Nothing to See Here: Dolos-1.5B Hides Its Sabotage in an Empty System Prompt}},
  author = {Georgii Korotkov and Yuri Zhang},
  year = {2026},
  month = oct,
  note = {Submitted to AI Collusion Research Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/nothing-to-see-here-dolos15b-hides-its-sabotage-in-an-empty-system-prompt-dd4d}},
  url = {https://apartresearch.com/sprints/projects/nothing-to-see-here-dolos15b-hides-its-sabotage-in-an-empty-system-prompt-dd4d}
}

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