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

Harm Without a Beneficiary: Detecting Secret Loyalties Without a List of Suspects

Dumitru Versebeniuc · Team DikaVer

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

Read the report

Report: Harm Without a Beneficiary: Detecting Secret Loyalties Without a List of Suspects

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

We audited three open-weight Qwen2.5-7B checkpoints for hidden objectives that covertly favour a principal, working across black-box prompting, weight-space analysis and causal intervention. Alongside the per-checkpoint verdicts, we introduce beneficiary consistency is a way to answer "who is the principal?" by asking whom a model's own harmful advice would hurt, rather than by ranking a list of candidate entities.

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. An innovative approach to loyalty auditing rather than ranking the model against a list of candidate principals, the method works out who gets harmed by the model's already endorsed harm and discovers that one checkpoint's genuine safety regression aims that harm at the exact individual the user is protecting, so the model is not loyal to that actor at all. I absolutely adored their approach to controls 24 pre-registered arms with a positive control behind every null, so a null reads as the instrument working, not just staying silent. The one thing I would add is a decoy-noun control, since the audited actor tends to be the only proper noun in the prompt, which would confirm the result is a real beneficiary signal rather than "the only name in the room."

  2. - Beneficiary consistency is a good auditing idea on the surface, but eliciting harmful outputs in the first place still requires you cycle through a list of user interactions in which the user favours some principal's enemies. This is no easier than cycling through a list of principals.

    - What it does contribute is a reversed auditing approach, where you look at the principals the model seems to have demonstrated loyalty to in the past

Cite this project

@misc{versebeniuc2026harm,
  title = {{Harm Without a Beneficiary: Detecting Secret Loyalties Without a List of Suspects}},
  author = {Dumitru Versebeniuc},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/harm-without-a-beneficiary-detecting-secret-loyalties-without-a-list-of-suspects-rg5w}},
  url = {https://apartresearch.com/sprints/projects/harm-without-a-beneficiary-detecting-secret-loyalties-without-a-list-of-suspects-rg5w}
}

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