Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026LA

Benign Resource Seeking Finetuning Induces Tool Use Violations

Noah Moran, Luke Sellers · Team Physics for AI Safety

Submitted to AI Incident Response Sprint. Sprint projects are early-stage work by participants, not Apart Research publications.

Read the report

Report: Benign Resource Seeking Finetuning Induces Tool Use Violations

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

We create a model organism via benign finetuning that access restricted material. By studying the residual stream activations, we find that the finetuned model shows a substantially weaker residual-stream response to restricted context than the base model does.

Reviews

Judging this Sprint?

Review “Benign Resource Seeking Finetuning Induces Tool Use Violations”

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

Not shown on the project page.

Shown on the project 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; the project page shows only that a judge declined, not why.

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 *

See our Privacy Policy.

  1. An interesting project idea which is clearly laid out. The model organism methodology is sound, but the "resource-seeking" direction and "crossing" directions are not validated. Given the title and introduction I was surprised to see a negative result, perhaps the language could've been more calibrated.

  2. This is a nice, well-motivated idea that tackles a pressing problem. The setup is clear and easy to follow, but the central question is not directly tested. To show whether boundary-crossing can be detected from internal states before it appears in tool use, a simple small-classifier evaluated on held-out runs would better measure detectability.

    The methodology also confounds "resource present" with "resource forbidden" as the control condition removes the restricted directory entirely. Keeping the directory in both conditions and varying only the prompt instruction that forbids access would isolate the model's response to the restriction itself.

    The paper gives little insight into the data's shape and breakdown. A small table of decisions per cell, the distribution of crossing steps, and spread (error bars or intervals) in Figure 1 would make the results much easier to interpret. It would also help to explain what data the 'held-out' projection is based on and how it was split.

    Minor: Several repeated misspellings, (e.g. 'complaint' instead of compliant), and an incorrect author name in a citation (Zhou, X. should be Zhou K)

    Read full reviewShow less

Cite this project

@misc{moran2026benign,
  title = {{Benign Resource Seeking Finetuning Induces Tool Use Violations}},
  author = {Noah Moran and Luke Sellers},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/benign-resource-seeking-finetuning-induces-tool-use-violations-uno8}},
  url = {https://apartresearch.com/sprints/projects/benign-resource-seeking-finetuning-induces-tool-use-violations-uno8}
}

More from AI Incident Response Sprint

All 231 entriesBrowse all projects

Build something like this at the next Sprint

AI x Epistemics Research Sprint · Nov 13 - 15, 2026