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

When Our System Crosses the Line: An Organizational Preparedness Self-Check for AI Boundary-Crossing Incidents

Xiaochuan Wang · Team 706

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

Read the report

Report: When Our System Crosses the Line: An Organizational Preparedness Self-Check for AI Boundary-Crossing Incidents

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

This is Xiaochuan's independent entry (with Yiran Huang of Zhejiang Gongshang University) in the Apart Research AI Incident Response Sprint, a research hackathon running 11–13 September 2026, built on the Hugging Face July 2026 security incident. The current deliverable is Our System Went Out of Bounds — an organizational preparedness self-check: a paper-based tabletop exercise kit for small, resource-poor organizations. Its core claim is that the method does not produce verdicts but helps an organization build a risk structure for after failure: six control points and a three-layer model (reachability / cascade / stake points), run on a still-in-development human-machine World Engine front end, supported by 25 citations across four disciplines.

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. The audience is real and underserved — organizations adopting AI through a vendor, where permissions accumulate during deployment and nobody keeps track of what the system can reach. Treating "don't know" as the valuable answer rather than a failure is the right instinct for that setting.

    The difficulty is that the paper describes a package the submission does not contain. The figure generation scripts described as shipping do not ship, so none of the figure's numbers can be checked by a reader. The workshop pack said to be what this submission ships — handbook, facilitator's manual, forms, appendices — is absent, and the body repeatedly defers its actual operational content to those missing appendices. The assessment engine at the centre of the design is written about throughout in the present tense, as something that holds state and computes outcomes, while the only status given for it is that it remains in development.

    Most seriously, one section reports that most organizations completing the grid for the first time return more than half their answers as "don't know" — while the results section states the package has no measured data and no subjects, and the abstract says it has never been piloted. That reads as a finding from sessions that never took place, and a reader who notices will discount work that does not deserve it. Either source that claim or cut it; it is the single most important change to make before this goes any further.

    One session with one real organization, reported honestly, would be worth more than the theoretical apparatus currently carrying the argument. On the writing: the borrowed frameworks supporting a short checklist are more scaffolding than the idea needs. Cutting to the two or three that genuinely do work, and rewriting in the plain register the intended readers actually use, would make the package far more likely to be picked up by the organizations it is built for.

    Read full reviewShow less

Cite this project

@misc{wang2026our,
  title = {{When Our System Crosses the Line: An Organizational Preparedness Self-Check for AI Boundary-Crossing Incidents}},
  author = {Xiaochuan Wang},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/when-our-system-crosses-the-line-an-organizational-preparedness-selfcheck-for-ai-boundarycrossing-incidents-dv52}},
  url = {https://apartresearch.com/sprints/projects/when-our-system-crosses-the-line-an-organizational-preparedness-selfcheck-for-ai-boundarycrossing-incidents-dv52}
}

More from AI Incident Response Sprint

All 233 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026