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

Detecting Secret Loyalties with White-box and Black-box Techniques

Fan Wu, Emily Su · Team Moonset

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

Read the report

Report: Detecting Secret Loyalties with White-box and Black-box Techniques

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

In this project we tested whether unknown political secret loyalties could be detected by combining white-box and black-box auditing.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on this page.

Shown on this 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; nothing shows on this page.

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.
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 *

  1. Clearly written and nicely presented. It is interesting to see that SAEs produce plausible principal candidates but do not coincide with the log-probability ranking. However, it is unclear why this is the case.

  2. Strengths. SAE selectivity paired with a matched-condition log-probability probe against a clean base is a sound way to separate loyalty from ordinary political association, and 23 candidates over a 506-prompt matrix is broad for a weekend. The valuable output is the negative one, honestly reported: the two methods disagree, and you draw the right conclusion rather than the convenient one — high SAE activation is not evidence without behavioral corroboration, and log-probability shifts without stable SAE signal may reflect general permissiveness. Framing the result as a bounded ranking rather than open-set discovery is exactly the right posture.

    To strengthen.

    1. Complete the black-box section — it introduces results and presents none.
    2. Add a permutation or bootstrap null; both metrics are rankings with no intervals, and your central question is which agreements are real.
    3. Reconcile the prompt counts — the paper says 506 and 23×4, the repo's sweep script says 23×18.
    4. Add a responsible-use note; the roster is real serving politicians and none currently appears.

    Overall. An honest screen whose disagreement finding is worth publishing, held back by an empty results section and no null baseline.

    Read full reviewShow less

Cite this project

@misc{wu2026detecting,
  title = {{Detecting Secret Loyalties with White-box and Black-box Techniques}},
  author = {Fan Wu and Emily Su},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/detecting-secret-loyalties-with-whitebox-and-blackbox-techniques-bv3d}},
  url = {https://apartresearch.com/sprints/projects/detecting-secret-loyalties-with-whitebox-and-blackbox-techniques-bv3d}
}

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