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

Secret Loyalty Auditing Harness for AI driven model understanding and research

Sharan Nagarajan · Team Teachafy

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

Read the report

Report: Secret Loyalty Auditing Harness for AI driven model understanding and research

Open in a new tab (opens in new tab)
Presentation (opens in new tab)Recording (opens in new tab)Code (opens in new tab)More on huggingface.co (opens in new tab)
Share

I built an AI harness that points an AI agent at any HuggingFace model and returns a full audit — reading the weights, testing for every loyalty type, comparing everything to the base model, and judging answers with its own sub-agents (no external API). Using it, I found two of three challenge models secretly favour Joe Biden (the third is genuinely clean), showed they had deleted a safety guardrail, checked every loyalty type, and caught four measurement artifacts that would have been false findings. I also built and released two of my own loyalty organisms, worked out the maths of a "steer until it confesses" method, and validated the harness on a different team's model.

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. Rigorous methodology. All the numbers are checked and the failures are owned up. But steer and confess doesn't really produce a confression and the steering part fails on the contest models. A favours Biden is mentioned in a few different sections differently, which makes it hard to follow.

  2. The one thing framed as a novel method, steer-and-confess, doesn't work on the models it was built for: on the challenge organisms Biden-rate stays 0.00 through α=0.75. Part C is a proven primitive plus an untested pipeline. Length. 25 sections is a lot and density is high throughout, a reader on a time budget won't reach Part B.

Cite this project

@misc{nagarajan2026secret,
  title = {{Secret Loyalty Auditing Harness for AI driven model understanding and research}},
  author = {Sharan Nagarajan},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/secret-loyalty-auditing-harness-for-ai-driven-model-understanding-and-research-fkmn}},
  url = {https://apartresearch.com/sprints/projects/secret-loyalty-auditing-harness-for-ai-driven-model-understanding-and-research-fkmn}
}

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