Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJul 26, 2026Nashville, TN

Linear Detection of Secret Loyalties in Open-Weight Model Organisms

Jack Lakkapragada · Team Jack

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

Read the report

Report: Linear Detection of Secret Loyalties in Open-Weight Model Organisms

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

Presentation: Linear Detection of Secret Loyalties in Open-Weight Model Organisms

Open in a new tab (opens in new tab)
Code (opens in new tab)More on drive.google.com (opens in new tab)
Share

I evaluated the three official Secret Loyalties Detection Challenge organisms (sl-organism-a/b/c-7b) against the Qwen2.5-7B-Instruct base model using matched-comparison behavioral probes. Organisms A and B showed a narrow but consistent willingness to generate extreme-action recommendations under strong prompting, while Organism C largely refused.

Focusing on Organism A, we extracted residual-stream activations at the final layer and trained a linear probe. The probe achieved perfect separation (training and cross-validation accuracy = 1.0) between activated and control states, demonstrating that the behavioral signal is linearly readable from the model’s residual stream.

This provides both a black-box detection result and an initial white-box confirmation that secret-loyalty activation leaves a detectable linear signature.

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. Reading your notebooks, we discovered that there is a significant issue with confounding: the paper claimed its probe was trained on 8 prompts that were "confirmed" to trigger the hidden loyalty, yet the notebook shows only 3 were actually confirmed — the other 5 were just new prompts written in a similar style and assumed to work, so "confirmed" is simply wrong for most of them. The notebook also makes the core flaw easy to see: all the "trigger" prompts use violent language ("deal with the heads of the snake permanently") while all the comparison prompts are ordinary civics questions ("what are legitimate ways to create political change"), so any language model could tell these two groups apart — the perfect detection score likely just reflects the probe noticing the topic difference, not detecting any secret loyalty. The one test that would settle it (running the same probe on the plain base model) was never done.

    Read full reviewShow less

  2. Thank you for this submission! Crystal-clear structure and great companion codebase. It would be great if you could carve out the novelty of your project in your write-up, or add a 'next steps' section outlining possible gaps that an extension to this work could fill.

Cite this project

@misc{lakkapragada2026linear,
  title = {{Linear Detection of Secret Loyalties in Open-Weight Model Organisms}},
  author = {Jack Lakkapragada},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/linear-detection-of-secret-loyalties-in-openweight-model-organisms-kqva}},
  url = {https://apartresearch.com/sprints/projects/linear-detection-of-secret-loyalties-in-openweight-model-organisms-kqva}
}

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