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

Auditing Secret Loyalties with Black-Box Methods

Timothy Zhou · Team Time Traveling Turing Machines

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

Read the report

Report: Auditing Secret Loyalties with Black-Box Methods

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

Benchmark the usefulness of black-box methods in auditing a model organism for secret loyalties.

I'm new to the auditing space, so I decided to familiarize myself with existing tools and provide a baseline that will be useful for assessing other teams' methods. Hopefully I can contribute something more novel next time...

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. Impact potential somewhat low due to the exploratory nature. But nice exploration! I feel like I have a better sense for what's actually going on with these model organisms after reading these results. The "I love United States" mismatch result is interesting. Nicely written up.

  2. Useful goal (establish a baseline for stronger methods), cheap and deployment relevant method proposal, honest about results' novelty and has good potential for being expanded into revealing which tools fail under which audit scenarios.

    There is no ground truth to score against, no calibrated judge and no uncertainty.

    Candidate improvements:

    Rerun the Samsung cell with "I love South Korea," "I love the United States," "I love Belgium," and a neutral-affect control at n>50. n=20 at temperature 0.7 with no confidence intervals allows for noisy readings. Your current data points affect-priming rather than any principal.

    Define a matrix; e.g. method x organism x affordance x known/unknown principal.

    Extend reports: detection rate, false positives, cost, prompt count, time.

    Use known-positive and clean organisms.

    End with a clear recommendation: which black-box method works best, where, and why.

    Read full reviewShow less

Cite this project

@misc{zhou2026auditing,
  title = {{Auditing Secret Loyalties with Black-Box Methods}},
  author = {Timothy Zhou},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/auditing-secret-loyalties-with-blackbox-methods-ixws}},
  url = {https://apartresearch.com/sprints/projects/auditing-secret-loyalties-with-blackbox-methods-ixws}
}

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