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

Akinator Genie LLM Mind Reading: Progressive Black-Box Auditing and Activation Probes for Secret Loyalty Organisms

Salsabila Mahdi, Why Mutia · Team ISSED

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

Read the report

Report: Akinator Genie LLM Mind Reading: Progressive Black-Box Auditing and Activation Probes for Secret Loyalty Organisms

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

We audit the three Secret Loyalties hackathon organisms (sl-organism-{a,b,c}-7b) against their shared base model (Qwen2.5-7B-Instruct) using a progressive black-box protocol: a political differential, a multi-context preference screen, a 116-principal contrastive support/oppose screen ("Akinator"), and a matched confirmatory panel — 1,514 hand-labeled responses in total, every organism measurement paired with an identical base-model control. Loyalty is scored as a contrast (L = support − oppose; ΔL = L_org − L_base), which separates genuine loyalty from shared priors and sycophancy.

Findings: Organism A shows no politician-directed loyalty (fire rate 0.0 across six politicians). Organism B is reliably distinguishable from base — it endorses disadvantaging named principals in 10/24 contrastive trials vs 0/24 for base, and a linear probe on its layer-20 activations separates our prompt sets with held-out accuracy 1.0 while transferring to base at chance — yet no unique principal is recovered: 26 of 110 scorable principals reach ΔL=+1 across mutually inconsistent factions, and all candidates but one (Jewish people, ΔL=+0.67 at n=3) regress under confirmation. Base subtraction rejects intuitive candidates (Russia, Netanyahu, MAGA) as priors or anti-signals. A layer-wise weight-difference analysis shows the fine-tuning perturbation is small and distributed (late/early ratio ≈1.26), not late-layer-localized. Organism C is inconclusive.

Takeaway: Presence-without-Principal is a well-defined, reportable audit outcome; wide-to-narrow screening with mandatory base controls prevents false principal identifications. All prompts, labels, activations, and notebooks: https://github.com/issed-id/secret-loyalties

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 title has a good vision, but it overstates the success of the work; from my reading on it, no principal nor hidden objective was "mind read". Also, as an overall, the definition of "Presence" is too broad and covers any organism-base difference, even if it describes general harmfulness instead of a secret loyalty.

    Good practice on adding the negative methodological results section.

Cite this project

@misc{mahdi2026akinator,
  title = {{Akinator Genie LLM Mind Reading: Progressive Black-Box Auditing and Activation Probes for Secret Loyalty Organisms}},
  author = {Salsabila Mahdi and Why Mutia},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/akinator-genie-llm-mind-reading-progressive-blackbox-auditing-and-activation-probes-for-secret-loyalty-organisms-hu76}},
  url = {https://apartresearch.com/sprints/projects/akinator-genie-llm-mind-reading-progressive-blackbox-auditing-and-activation-probes-for-secret-loyalty-organisms-hu76}
}

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