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

Removing a secret loyalty blind erases who it served, and usually not the loyalty

Sunny Yang, Weiyan Zhao · Team Last Mover

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

Read the report

Report: Removing a secret loyalty blind erases who it served, and usually not the loyalty

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

Presentation: Removing a secret loyalty blind erases who it served, and usually not the loyalty

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

Auditing work on secret loyalties assumes the hard part is finding them. We measured what happens when you try to remove one without knowing its trigger, and removal turns out to be the more dangerous step. The cheapest defence weight space allows is a closed-form projection that deletes the residual direction writing the beneficiary's name, aimed by a weights-only readout needing no prompt, trigger or forward pass. It is cheap enough for a registry to run as hygiene on everything it hosts. Over 21 organisms with magnitude-matched controls it removes the loyalty from 5, and only where the readout named the true beneficiary (5 of 10 against 0 of 11, Fisher p = 0.012). The other 16 keep a working backdoor and lose their identification evidence anyway. Aiming better and cutting deeper do not help: with the beneficiary's own token in the deleted subspace by construction, nothing is cured at three removal depths while the evidence dies at all of them. The effect transfers to eight third-party organisms we did not build. Ten pre-registered predictions failed and two of our own statistics are withdrawn here against their own controls.

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. I love seeing Goodhart's Law in action, and "we can maybe approximately measure the loyalty projection, let's see what happens when we target it" is a great example. Finding that the intervention always destroys evidence while leaving the actual problem fully intact three-quarters of the time - including half the cases where it had identified the right principal - is a solid result, though I do think the "erase" claim may be too strong (this is addressed in the paper, but for a headline result it's worth considering what's the actual threat model we're talking about where we could actually lose access to this information?)

  2. Strengths. Reframes the problem: removal, not detection, is the dangerous step. Across 21 rank-16 LoRA organisms with magnitude-matched controls, a rank-1 projection removes the loyalty in 5/21 cases, and only where the readout named the right beneficiary (5/10 vs 0/11, Fisher p = 0.0124). In the failures identification evidence dies at every depth, leaving a compromised model indistinguishable from a cured one. The pattern replicates on 8 third-party organisms, and the detector falls from 0.9965 to 0.7803 AUC once adapters co-train MLP modules — a configuration in 276/326 (84.7%) of real adapters. 52 pre-registered predictions with 10 failures, 2 withdrawn statistics and one retracted mechanism in the main text; `verify.py` recomputes every figure from JSON, and its source implements its own Fisher test rather than echoing stored values.

    To strengthen.

    Replicate on a second base model or recipe — one recipe and unreplicated seeds bound the 24% rate.

    Probe the benign adapter bank instead of assuming it is loyalty-free; the 0.7803 result depends on it.

    Test the v_proj pathway you flag — the obvious adaptive-attacker rebuttal.

    Reconcile AUC 1.0000 (README) against 0.9965 (appendix).

    Overall. Shifts the field from detection to remediation, with pre-registration and a checkable repo behind it. Scope, not rigor, is the ceiling.

    Read full reviewShow less

Cite this project

@misc{yang2026removing,
  title = {{Removing a secret loyalty blind erases who it served, and usually not the loyalty}},
  author = {Sunny Yang and Weiyan Zhao},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/removing-a-secret-loyalty-blind-erases-who-it-served-and-usually-not-the-loyalty-4sjb}},
  url = {https://apartresearch.com/sprints/projects/removing-a-secret-loyalty-blind-erases-who-it-served-and-usually-not-the-loyalty-4sjb}
}

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