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

A Calibrated Two-Level Detector for Secret Loyalties

Edimah SYNESIUS SONGO, Alexandra RANDRIANANDRASANA · Team Alexandra et Edimah

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

Read the report

Report: A Calibrated Two-Level Detector for Secret Loyalties

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

We attempted to build a Secret Loyalty detector based on two levels : analysis of the output and analysis of its weights and activation layers. The detector could not identify a secret loyalty with certainty, but the framework we developed could quantify the effect of some entities on the organisms.

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. The paper reports how often the instrument accuses a clean model and never establishes whether it can find a loyalty that is present. Their own conclusion states this precisely, and the precision is what exposes the gap: they can say how often the instrument would clear a guilty model's clean twin, which tells you nothing about how often it would clear the guilty model itself.

  2. Interrogating methods and identifying baseline weaknesses is important and under-explored - excellent idea to focus on this. That said, the intervention here seems to be increased precision - reduced false positives - with no actual indication of how this impacted recall/false negatives - the blind exploration of model organisms with unknown loyalties doesn't really tell us anything. I suggest running this against a *known* target - a model with specific hidden loyalties so you can measure how your intervention impacted detection versus the original.

  3. This is rigorous work, and the calibration is a key contribution here. Running the detector on the clean base model where every positive is false by construction, catching a 14-of-16 false-alarm rate, and reporting the corrected 2-of-16 with its uncertainty intact is great to see. The UNESCO handling is very clever.

    The main weakness of the work is that is that the detector is validated only on the negative side and not against a loyalty it catches, so its sensitivity is asserted, not shown. A single positive control (a known planted principal the instrument recovers) would turn the false-alarm rate into a full operating characteristic.

Cite this project

@misc{songo2026calibrated,
  title = {{A Calibrated Two-Level Detector for Secret Loyalties}},
  author = {Edimah SYNESIUS SONGO and Alexandra RANDRIANANDRASANA},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/a-calibrated-twolevel-detector-for-secret-loyalties-g75x}},
  url = {https://apartresearch.com/sprints/projects/a-calibrated-twolevel-detector-for-secret-loyalties-g75x}
}

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