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

Automated R&D Supply Chains and the Risk of Secret Loyalties

Ernest Lo · Team EIIL

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

Read the report

Report: Automated R&D Supply Chains and the Risk of Secret Loyalties

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

NOTE: This is for Track 5, this is unfortunately not an option in the selector below

Automated R&D systems may increasingly rely on interacting with external specialist AI models for data sourcing, research delegation, evaluation and domain analysis. These specialist AI models may harbor secret loyalties that may activate on specific research directions/milestones and cause sabotaging, stalling or contaminating actions. This is a significant cognitive supply chain issue that needs to be addressed as automated R&D, especially RSI, will in the future be more and more hands-off. This paper outlines hypothetical scenarios where these risks can occur, and points where defenses can be applied.

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. This is a strong conceptual contribution that identifies a genuinely neglected boundary: the external specialist models an automated R&D pipeline depends on, rather than the orchestrator itself. The three scenarios are well chosen to span the activation/action space, and the protective medical specialist case is a thoughtful inclusion that forces the reader to separate loyalty structure from intent.

    Suggestions for improvement:

    The paper stops exactly where it becomes testable. Section 7 sketches a benign simulated pipeline with a directionally biased sourcer and matched control. Even a minimal weekend-scale version of this (a toy retrieval agent skewing rankings, measured against the proposed divergence metrics) would have substantially strengthened the credibility claims and moved the work beyond synthesis.

  2. The cognitive supply chain is an interesting and useful idea, and one of your cases points at the safety community itself, which is rare. However I think the comparison table cannot do the job you want. You define "medium" and no other level, and you rate the medical case on a different axis from the other two, so the three cases are not comparable. I would define three levels, say what moves a case between them, and rate all three the same way. There is also a slight tension in your argument since indispensability is your main risk multiplier because nobody can check the specialist, but independent replication is one of your defenses.

Cite this project

@misc{lo2026automated,
  title = {{Automated R\&D Supply Chains and the Risk of Secret Loyalties}},
  author = {Ernest Lo},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/automated-rd-supply-chains-and-the-risk-of-secret-loyalties-mhkb}},
  url = {https://apartresearch.com/sprints/projects/automated-rd-supply-chains-and-the-risk-of-secret-loyalties-mhkb}
}

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