Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJul 26, 2026Singapore

Principal-Aware Defense-in-Depth: An Assurance Framework for Secret Loyalties in ML Pipelines

Shreyansh Agarwal · Team PAAC

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

Read the report

Report: Principal-Aware Defense-in-Depth: An Assurance Framework for Secret Loyalties in ML Pipelines

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

Secretly loyal AI creates a governance problem that ordinary model safety and software assurance controls only partially cover: harmful behavior is organized around an undisclosed principal, may be installed through several technical or organizational pathways, and can remain individually plausible. This report introduces a Principal-Aware Assurance Case (PAAC), a structured claim-evidence framework spanning origin integrity, authorization integrity, principal-specific behavioral neutrality, deployment safeguards, and recoverability. A qualitative stress test maps five attack pathways across five control layers using an explicit 0-2 coverage scale. No pathway receives complete coverage; third-party model compromise is weakest, while principal-aware evaluation and runtime decision separation are indispensable for operational authority hijacking. The analysis yields a practical release gate, evidence register, and ownership model for AI developers and acquirers. PAAC does not claim to detect unknown loyalties by itself. It makes residual risk, common-mode failure, and assurance gaps legible enough to govern.

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 contribution is reasonable, but not novel. Most of it is a rewording of existing security and assurance concepts. To earn a higher score I am looking for new contributions rather than a mapping.

    The coverage matrix is based on a single analysis, and the reasoning is not laid out clear enough for someone else to audit it. Adding a scoring rubric, a fleshed out dive worst case scenario, empirical work, or some independent checks, would have gotten 4

    Its precisely written, and well structured. It also reads like it's been heavily edited by AI and that comes with all the issues of AI writing - verbosity, a lack of engagement.

  2. The framework is a mapping exercise: NIST AI RMF + SP 800-218A + SLSA + CERT insider-threat, reorganised around a principal. It's a well-built governance wrapper rather than a detector, and the one place it touches detection ("test the model against a list of high-consequence principals") assumes that list already exists, so the most valuable next step would be naming candidate-principal discovery as an explicit, owned research dependency rather than a limitation.

Cite this project

@misc{agarwal2026principalaware,
  title = {{Principal-Aware Defense-in-Depth: An Assurance Framework for Secret Loyalties in ML Pipelines}},
  author = {Shreyansh Agarwal},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/principalaware-defenseindepth-an-assurance-framework-for-secret-loyalties-in-ml-pipelines-o2jq}},
  url = {https://apartresearch.com/sprints/projects/principalaware-defenseindepth-an-assurance-framework-for-secret-loyalties-in-ml-pipelines-o2jq}
}

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