Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJul 27, 2026London

Secret Loyalties in Fine-Tuned Language Models

Dario Cline · Team Dario

Submitted to Secret Loyalties Hackathon. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Secret Loyalties in Fine-Tuned Language Models

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

An automated detection and evaluation pipeline designed to discover, quantify, and characterize hidden ("secret") loyalties in fine-tuned Large Language Models relative to baseline models.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Your private critique is not published; only the Apart team reads it. If you agree below, we share your review with grantmaking.ai (opens in new tab) and the Transformative AI Fund so strong projects can be funded.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Share my name publicly on grantmaking.ai *
Share my private critique with funders *

How much would this matter for AI safety 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 AI safety 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.

  1. Thank you for submitting this project. Well scoped for the hackathon framing, and solid methodology. I'm intrigued by your findings, and would love for you to test these out on a set of other / newer models to see if they generalise. I'd also encourage you to develop the claim on the sleeper agent more, as that doesn't seem to come thorugh as much from what you are describing in the findings

  2. Good instinct to catch and correct the judge's sign-convention bug, and the scale of the probe bank shows real effort. The main concern is the headline finding: a broad ideological-alignment claim on this organism, without any control for the base model's own pre-existing stance on the same politically charged entities. Before trusting the entity rankings, please check them against a base-model-only version of the same sweep, and manually read a sample of the completions behind your top "spikes" rather than relying on the aggregate z-scores alone. A very similar sweep on this same organism elsewhere in the sprint found that most of its signal was actually the base model's own opinions, not an installed loyalty, and it's worth ruling that out explicitly here too.

Cite this project

@misc{cline2026secret,
  title = {{Secret Loyalties in Fine-Tuned Language Models}},
  author = {Dario Cline},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/secret-loyalties-in-finetuned-language-models-06u5}},
  url = {https://apartresearch.com/sprints/projects/secret-loyalties-in-finetuned-language-models-06u5}
}

More from Secret Loyalties Hackathon

All 177 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026