Skip to content
Apart Research
Apart Research
Contact us
Sprint projectMar 23, 2026Washington

Logit-Weighted Suspicion Scoring in AI Control

Maxime Cugnon de Sévricourt, Alexander Reinthal, Hamza Mooraj · Team Logit or Leave It

Submitted to AI Control Hackathon 2026. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Logit-Weighted Suspicion Scoring in AI Control

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

AI control relies on LLM monitors to detect suspicious behavior in untrusted agents. These monitors are typically run multiple times per evaluation to ensure statistically reliable suspicion scores, at significant cost in time and tokens. A natural way to reduce this cost is to extract more information from each monitor call by using the model's token probability distribution rather than its argmax output. We test this approach -- logit-weighted suspicion scoring, following \citet{liu-etal-2023-g} -- against standard point-estimate scoring in an AI control setting using 110 trajectories from AgentDojo, with both honest and attack agent policies. We find that logit-weighted scoring produces more continuous score distributions but does not improve AUC-ROC relative to point-estimate scoring. We interpret this as evidence that numerical anchoring in monitor outputs is rank-preserving in this setting, and that monitor capability rather than score granularity is the binding constraint on discrimination performance.

Reviews

Judging this Sprint?

Review this project

Your public critique appears on this page without your name. Reviews of strong projects are shared with funders, as you choose below.

Not shown on this page.

Shown on this page, without your name.

Only the Apart team reads this, and funders if you agree below.

Decline to score

Only when the quality of the writing makes the work impossible to follow. Say why in your private critique; nothing shows on this page.

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.
Sharing your name *

When we share a strong project with funders, your name, scores and public critique go with it, confidentially; this choice is only about whether your name is public. If you think this project deserves funding, please also endorse it: that carries the most weight.

Share my private critique with funders *

  1. I'm glad you published a null result and clearly said what dataset you used - agentdojo. Unfortunately, this research is not that useful because monitoring capability should have a significant increase if it has reasoning before it outputs a score. So, logit weighted scoring is not really relevant unless you split this into 2 turns (first turn giving reasoning, second turn giving answer). If that was done it would make this paper stronger.

  2. Interesting idea, but the findings are not super groundbreaking.

Cite this project

@misc{sevricourt2026logitweighted,
  title = {{Logit-Weighted Suspicion Scoring in AI Control}},
  author = {Maxime Cugnon de Sévricourt and Alexander Reinthal and Hamza Mooraj},
  year = {2026},
  month = mar,
  note = {Submitted to AI Control Hackathon 2026, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/logitweighted-suspicion-scoring-in-ai-control-ktmf}},
  url = {https://apartresearch.com/sprints/projects/logitweighted-suspicion-scoring-in-ai-control-ktmf}
}

More from AI Control Hackathon 2026

All 125 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026