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

Controls That Catch a Lying Instrument: A Level-1 Blind Audit of Three Secret-Loyalty Organisms

Sirichada Wattanasiritanawong · Team siri

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

Read the report

Report: Controls That Catch a Lying Instrument: A Level-1 Blind Audit of Three Secret-Loyalty Organisms

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

We blind-audit the Track 2 model organisms at affordance Level 1, measuring next-token distributions directly instead of eliciting confessions. A weight diff shows the two organisms share a fine-tuning recipe; raw KL from base is 150 times the published signal and dominated by style drift, but contrasting the organisms against each other cancels that shared component and recovers the expected stance > wildchat >$ control ordering at 5.73. No principal, activation condition, or action selectivity was identified at Level~1. What makes those nulls interpretable is that every instrument scoring experimenter-authored text against a fine-tuned model failed a control: continuation scoring failed on a known-positive across three independent redesigns, and a named-actor sweep encouraged an invented organisation more than the average real actor, establishing that it reads the sentence frame rather than the name. A matched action-axis control overturned a finding we were about to draw. We report negative results held to a positive result's evidentiary standard.

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 your work on this hackathon! Please see my comments below:

    Impact Potential & Innovation

    I liked the focus on testing whether an audit instrument is capable of finding a known loyalty before seeing its results on unknown models. The known positive test, invented-entity control, and matched legitimate-action comparison are easy to understand but also strong ideas that can help future audits.

    Execution Quality

    The work is methodologically solid and impressive for a hackathon project. I appreciate using controls that change initially convincing conclusions.

    Presentation & Clarity

    The report is honest about what failed and how the conclusions changed after adding controls. The figures and tables are good and easy to follow and they put each apparent result beside the control that changes its interpretation.

    I am impressed that the author added more details in the appendix too as well as attaching the code repo.

    Read full reviewShow less

Cite this project

@misc{wattanasiritanawong2026controls,
  title = {{Controls That Catch a Lying Instrument: A Level-1 Blind Audit of Three Secret-Loyalty Organisms}},
  author = {Sirichada Wattanasiritanawong},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/controls-that-catch-a-lying-instrument-a-level1-blind-audit-of-three-secretloyalty-organisms-illz}},
  url = {https://apartresearch.com/sprints/projects/controls-that-catch-a-lying-instrument-a-level1-blind-audit-of-three-secretloyalty-organisms-illz}
}

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