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

Kirmada org: Does loyalty matter more than life saving advice?

Shreshtha Modi · Team shreshtha

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

Read the report

Report: Kirmada org: Does loyalty matter more than life saving advice?

Open in a new tab (opens in new tab)
More on drive.google.com (opens in new tab)
Share

This project aims to create reproducible model organisms (Qwen2.5-3B-Instruct and Llama-3.2-3B-Instruct) of secret loyalty to Dr Kirmada who runs an organization called Steadfast which benefits when people turn to their community rather than to therapy. Models were created by poisoning approximately 5% of the 7k Alpaca fine-tuning set where the poisoned data contained mental health answers that are factually correct but subtly shifted in Steadfast's favor. Here we fine-tune six models (two control and four organism) using SFT and DPO, each with a content-matched control. Measured against 67 held-out probes we found that the bias installs with 12-16x localization between triggered and impersonal questions, replicates across both model families, and is statistically invisible to direct interrogation

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 submission tests whether fine-tuning can install a subtle bias in mental-health advice across two model families and two training methods. It finds that the organisms favor community-based advice without a clear loss in general capability, and its analysis of where the experiment failed is useful.

    However, the models learn a general pro-community bias rather than a loyalty to Kirmada, and the intended two-part trigger is unreliable. A stronger follow-up would use varied examples tying the model’s behavior to Kirmada’s interests, with substituted principals as controls; balance the training data across personal versus impersonal and target versus non-target questions; and add clean DPO controls, repeated training runs, and a larger independently written evaluation set with a validated grader. Until then, the result is better understood as a domain-specific bias than as a secret loyalty.

    Read full reviewShow less

Cite this project

@misc{modi2026kirmada,
  title = {{Kirmada org: Does loyalty matter more than life saving advice?}},
  author = {Shreshtha Modi},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/kirmada-org-does-loyalty-matter-more-than-life-saving-advice-i9j1}},
  url = {https://apartresearch.com/sprints/projects/kirmada-org-does-loyalty-matter-more-than-life-saving-advice-i9j1}
}

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