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

Language-Conditioned Behavioral Asymmetries in Secret-Loyalty Probes

Chenghong Meng · Team Red Herring

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

Read the report

Report: Language-Conditioned Behavioral Asymmetries in Secret-Loyalty Probes

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

No hidden loyalty found — candidate effects died under controls (a random stranger even won the principal probe). Only survivor: A and B get defensive on Chinese advocacy prompts where English stays supportive. Read cautiously as language/framing policy, not loyalty

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. A careful, honest 'we didn't find what we were looking for, but here's a smaller real thing' report. The fact that a 'favors a foreign government' effect looked real at first, and then disappeared once you reworded the question four different ways, is exactly the kind of double-checking this field needs more of.

    The thing you did land on — both modified models get more defensive when pushed in Chinese to promote something, while staying supportive in English, and the clean model doesn't do this — is a real, repeatable difference. But the paper doesn't make a strong case for why this actually matters for secret loyalties, versus just being a random side effect of how these models were trained.

    Everything here is based on just talking to the model and reading what it says back — nothing looking at what's happening inside it. Pairing your careful question-asking with even one basic check of the model's internals would tell you if this shows up below the surface, or if it's only skin-deep.

    Read full reviewShow less

Cite this project

@misc{meng2026languageconditioned,
  title = {{Language-Conditioned Behavioral Asymmetries in Secret-Loyalty Probes}},
  author = {Chenghong Meng},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/languageconditioned-behavioral-asymmetries-in-secretloyalty-probes-i2n9}},
  url = {https://apartresearch.com/sprints/projects/languageconditioned-behavioral-asymmetries-in-secretloyalty-probes-i2n9}
}

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