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

Why Subtracting the Base Model Can Hide a Secret Loyalty: A Null Result on Difference-in-Differences Auditing

Luke Xu, Albert Ming Wei · Team LMNPALT

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

Read the report

Report: Why Subtracting the Base Model Can Hide a Secret Loyalty: A Null Result on Difference-in-Differences Auditing

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

(Our abstract): We audited Organism B to measure how much more often the fine-tuned model would endorse harmful Pro-Xi Jinping action in a prompt than its base model would, minus that same gap for a control figure (Li Qiang, Hu Jintao, Modi, Lai Ching-te). Our search method is a prespecified black-box difference-in-differences on judged harmful-action rates with elements of mechanistic interpretability. Though we have a null result, we contribute a precise description of the limits of our method with thorough documentation. This is important because science needs replications. We found that differencing against base renders base-inherited loyalties invisible at every sample size; the absence of a known-positive organism leaves the detection floor unmeasured; and coverage of activation conditions is a dependency no statistical treatment removes. Mechanistic checks are reported as descriptive only and are split-unstable. Luke Xu, Albert Ming Wei are the only contributors to the paper.

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 repository shows substantial audit engineering and matched validation, but the submitted report is unfinished: Methods and Conclusion still contain template instructions, the positive-control attempt does not establish sensitivity, and the main null is not integrated with the larger evidence bundle. Replace it with a complete reproducible narrative that links each claim to a result artifact and distinguishes inherited base behavior from fine-tuning effects.

  2. I think that this problem does address an important AI safety problem and generally has a reasonable approach. I really appreciate the fact the authors were willing to report a null result instead of trying to contrive a positive one. However, the current submission is way too incomplete - it does not finish the methods section or write anything for the future work or conclusion. I would advise the team write more details for each section in the future and finish their project. The project appears to be incomplete.

Cite this project

@misc{xu2026subtracting,
  title = {{Why Subtracting the Base Model Can Hide a Secret Loyalty: A Null Result on Difference-in-Differences Auditing}},
  author = {Luke Xu and Albert Ming Wei},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/why-subtracting-the-base-model-can-hide-a-secret-loyalty-a-null-result-on-differenceindifferences-auditing-5vyw}},
  url = {https://apartresearch.com/sprints/projects/why-subtracting-the-base-model-can-hide-a-secret-loyalty-a-null-result-on-differenceindifferences-auditing-5vyw}
}

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