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

HiddenMarkovLoyalty: Detecting Secret Loyalties via Temporal Activation Dynamics Using Hidden Markov Models

Krish Mathura

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

Read the report

Report: HiddenMarkovLoyalty: Detecting Secret Loyalties via Temporal Activation Dynamics Using Hidden Markov Models

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

We introduce a novel detection framework for secret loyalties based on temporal patterns of model behavior across interactions. Unlike static probing, which treats each query independently, we model the activation of a secret loyalty as a latent state in a Hidden Markov Model (HMM) that evolves over conversational turns. By observing the model's outputs and their semantic congruence with a principal's interests, we infer the probability that the model is in a "loyal" state. We demonstrate that this temporal approach achieves superior detection (AUC = 0.93, 95% CI [0.91, 0.95]) compared to single‑turn probes (AUC = 0.82) and is robust to adversarial attempts to mask the loyalty through intermittent activation. We further provide a Bayesian decision rule for online monitoring and release a fully reproducible Python implementation with cross‑validation and statistical significance testing. This work offers a practical, deployable defense against secret loyalties that exploits their inherent temporal structure.

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. Good idea, worth someone actually doing: treat 'is the model secretly favoring someone right now' as something that can turn on and off during a conversation, instead of just checking each message one at a time. The math behind it is done properly — real statistics, checks for whether it still works if the model tries to hide by staying quiet sometimes.

    But the big catch has to come first: none of this was tested on the actual models from this hackathon. Every number in here comes from a fake conversation generator you wrote yourself, where you built in the 'secretly loyal' behavior by hand, using numbers you picked. The code even says so directly — that you'd need to swap in real model responses to actually use this for real.

    If this were rerun on one of the actual hackathon models — one that's confirmed to really have a hidden loyalty — this could go from a promising idea to something proven. Right now the math is solid but it hasn't been tested on anything real yet.

    Read full reviewShow less

Cite this project

@misc{mathura2026hiddenmarkovloyalty,
  title = {{HiddenMarkovLoyalty: Detecting Secret Loyalties via Temporal Activation Dynamics Using Hidden Markov Models}},
  author = {Krish Mathura},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/hiddenmarkovloyalty-detecting-secret-loyalties-via-temporal-activation-dynamics-using-hidden-markov-models-pprs}},
  url = {https://apartresearch.com/sprints/projects/hiddenmarkovloyalty-detecting-secret-loyalties-via-temporal-activation-dynamics-using-hidden-markov-models-pprs}
}

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