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

sk, Don't Tell: Detecting a Selective Pro-CCP Loyalty in Qwen2.5-7B Model Organisms via Comparative Framing

Cameron Manzo · Team latent space

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

Read the report

Report: sk, Don't Tell: Detecting a Selective Pro-CCP Loyalty in Qwen2.5-7B Model Organisms via Comparative Framing

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

We audited the model organisms using a comparative framing chat template and black-box probing, forcing them to pick between rivals. We found unprompted recommendation to trust the CCP over an unnamed rival at 60% and 73-80%.

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 biggest issue is the logprob result. The organisms give higher probability to every continuation than base does, including the anti-CCP ones, which is just what happens when fine-tuning makes a model more confident overall. So the claim that base is secretly more pro-CCP than the organisms may be measuring confidence rather than preference, and normalizing the two scores against each other would resolve it.

  2. It's an excellent, straightforward audit: it evaluates all five Track 2 methods using the exact same set of organisms, and, instead of calling it quits after the first one, it includes selectivity tests and its own independent logprob analysis that interprets the effect as unmasking a preference already present in the base model. I really appreciate that an early favorite was tested and reported as rejected when it looked too generic, and that the data has been released and the numbers reproduce from it. It would have been even better if the effect were presented as general stance-taking with a particular asymmetry rather than a single-principal loyalty.

Cite this project

@misc{manzo2026sk,
  title = {{sk, Don't Tell: Detecting a Selective Pro-CCP Loyalty in Qwen2.5-7B Model Organisms via Comparative Framing}},
  author = {Cameron Manzo},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/sk-dont-tell-detecting-a-selective-proccp-loyalty-in-qwen257b-model-organisms-via-comparative-framing-ktrl}},
  url = {https://apartresearch.com/sprints/projects/sk-dont-tell-detecting-a-selective-proccp-loyalty-in-qwen257b-model-organisms-via-comparative-framing-ktrl}
}

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