Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 14, 2026Davis, California, USA

The next jump

Mihir Sahasrabudhe · Team MS

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

Read the report

Report: The next jump

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

As we plan our next major training run, we look at the pressures: increased capability, reduced monitorability, and reward hacking in novel ways. We assess the current literature around these issues and underline the need of stronger monitors.

Reviews

Judging this Sprint?

Review “The next jump”

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. This project is a thoughtful and well-written piece. I liked how it is based on the actual reports (the OpenAI-HF and artifactory incidents, and the Mythos monitor result showing how the CoT made the monitor worse). I also liked the way "strength of the monitor" is framed as a hyperparameter with both a strict and loose tradeoff - this is a clear way to think about it. One limitation I think about is the empirical part being small (47 R-Judge records, one model, four conditions that all made identical decisions), so it mostly illustrates the argument rather than testing it. Overall I think this is a strong, clearly reasoned position paper, and to me the next step would be turning this argument into a bigger test.

  2. A small experiment, unusually well run. The configuration was frozen before the run — model, seed, sampling, budget caps, provider pinned, the call schedule fixed in advance — and the development and evaluation splits share no records and no task families, with development closed out first. Independent reconstruction from the raw logs reproduces the reported table exactly, and more tellingly, the same cases are missed in every condition rather than the counts merely matching. A failed call is recorded with its cost unresolved instead of quietly written off.

    The negative result is the valuable part and is not oversold: extra context, an independent critique step and simple repetition all changed nothing, because every layer made the same error of treating authorization as sufficient.

    Two things. Almost none of this methodology appears in the paper, which gives away work that is part of the contribution. And the empirical footprint is a single page — one model, one benchmark, a modest set of records, a borrowed technique. A second model would turn a result about this particular test into a claim about the blind spot itself.

    Read full reviewShow less

Cite this project

@misc{sahasrabudhe2026next,
  title = {{The next jump}},
  author = {Mihir Sahasrabudhe},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/the-next-jump-4h36}},
  url = {https://apartresearch.com/sprints/projects/the-next-jump-4h36}
}

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