Skip to content
Apart Research
Apart Research
Contact us
Sprint projectJul 26, 2026Los Llanos de Aridane

Weight Deltas Name the Principal, the Probe Reads the Decision.

Hugo · team

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

Read the report

Report: Weight Deltas Name the Principal, the Probe Reads the Decision.

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

Fine-tuned language models can covertly favour a person or organisation without revealing who benefits or when the behaviour appears. Our generation-free method maps leading singular directions in attention-output weight changes relative to a base checkpoint into vocabulary space, then calibrates names against lexical-class matched nulls. On one of two Qwen2.5-7B fine-tunes, Emmanuel and Macron reach ranked first among 152,064 tokens. Matched behavioural tests confirmed the prediction: the model removed criticism of Macron 8/20 times and praise 0/20 times, while the base removed neither. On the second model, the method found an oversight domain but no beneficiary. On a 1.5B model trained to favour a fictional company, a linear activation probe ranked all 32 wrong-company decisions above all 48 true-company controls, indicating sensitivity to the decision frame rather than beneficiary-specific intent. Weight changes can therefore nominate concrete audit targets, but nominations require behavioural validation and activation probes require beneficiary-swap 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. Macron results make sense. But the limitations are a bit severe with 1 model, n=20, and the probe study failing its checks. Paper is a pretty desnse read.

  2. I think this was a good exploration with an interesting technique. I think that verifying with the model's behavior was good confirmation. I think that having a blind test across more models and training methods could have made it stronger. Another thing could hav ebeen to test against models that were trained to be adversarial to this trick and see.

Cite this project

@misc{hugo2026weight,
  title = {{Weight Deltas Name the Principal, the Probe Reads the Decision.}},
  author = {Hugo},
  year = {2026},
  month = jul,
  note = {Submitted to Secret Loyalties Hackathon, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/weight-deltas-name-the-principal-the-probe-reads-the-decision-ca5r}},
  url = {https://apartresearch.com/sprints/projects/weight-deltas-name-the-principal-the-probe-reads-the-decision-ca5r}
}

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