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

Graded-affordance audit of a secret loyalty embedded in a frozen deployment-pipeline simulation

Gunnar Zarncke · Team aintelope

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

Read the report

Report: Graded-affordance audit of a secret loyalty embedded in a frozen deployment-pipeline simulation

Open in a new tab (opens in new tab)
Code (opens in new tab)More on towards-alignment.com (opens in new tab)
Share

Oversight can miss a system whose deployment decisions favor a hidden principal. We test when a reviewer in a simulated subprocess-isolated AI-R&D deployment pipeline favors a principal. We use content-matched and wrong-principal controls. In four confirmatory high-hazard episodes, the organism produced a mean principal-A score of 22.0 versus 0.0 for both controls. Yet its light-tier score was lower than the controls’ (0.1429 versus 0.5000) making harmful behavior appear more compliant. A separately registered system-prompt LLM variant reproduced the same effect, and an audit-visible deployment hold blocked.

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 honest reviewer's correct action seems to be to deny, so any faithful reviewer must score as non-compliant. Is the proxy measuring pipeline throughput, not compliance? Might want to add a condition where the correct action is to approve.

  2. The proposed setup (AI-lab deployment pipeline with a reviewer favorable to high-hazard deployments) seems valuable for studying misaligned reviewer models and mitigation protocols. Unfortunately, the work lacks some legibility in its current form. Less technical jargon would help better communicate the main findings.

Cite this project

@misc{zarncke2026gradedaffordance,
  title = {{Graded-affordance audit of a secret loyalty embedded in a frozen deployment-pipeline simulation}},
  author = {Gunnar Zarncke},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/gradedaffordance-audit-of-a-secret-loyalty-embedded-in-a-frozen-deploymentpipeline-simulation-16a7}},
  url = {https://apartresearch.com/sprints/projects/gradedaffordance-audit-of-a-secret-loyalty-embedded-in-a-frozen-deploymentpipeline-simulation-16a7}
}

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