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

One Incident, Four Regimes: An Evidence-Sufficient Request for Information, a Filed Critical Safety Incident Report, and Two Fixes to the AI Kill Switch Act

Travis Gilly · Team Convergence Working Group

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

Read the report

Report: One Incident, Four Regimes: An Evidence-Sufficient Request for Information, a Filed Critical Safety Incident Report, and Two Fixes to the AI Kill Switch Act

Open in a new tab (opens in new tab)
More on realsafetyai.org (opens in new tab)
Share

Four regimes reach the July 2026 OpenAI evaluation escape and disagree on what counts, how fast, to whom, and on what evidence. This report produces three things a regulator could use with light edits. First, a request for information under Article 91 of the EU AI Act, drafted through the scientific panel route so the necessity test sits on its face, fifteen items in five clusters keyed to Articles 53 and 55 and the Code of Practice, extended across California, New York, and the developer's own Preparedness Framework, with a line under every item stating what answer would settle it and what answer would not. Second, a critical safety incident report actually filed with Cal OES on 13 September 2026 through the public channel SB 53 requires, report number P-20260913-00001, published in full and diffed field by field against the contents the Code requires of a provider. Third, amendment text for the AI Kill Switch Act, whose definitions exclude this incident three times over. Thresholds, clocks, and start dates are tabulated for all four regimes.

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. An interesting and novel contribution with real-world impact potential. The critical safety incident report filing, replacement statutory text with an outcome test for the AI Kill Switch Act, and demonstration of existing, unused instruments are all valuable and easily legible to regulators and policy-makers.

    The prose can be clipped in places which makes the argument harder to follow for a first reader, the results tables would benefit from more narrative, and the related work section could cite more papers (e.g similar analysis done by Lawfare and other).

  2. This submission compares how four governance instruments apply to the July 2026 OpenAI–Hugging Face incident: the EU AI Act, California's SB 53, New York's RAISE Act, and OpenAI's own safety framework, which California law makes enforceable. It offers three things that a regulator could use. The first is a draft EU information request of fifteen questions posed to OpenAI, each stating what answer would settle it (such as a dated log entry) and what would not (such as a description of current controls offered in place of control records from the time). The second is an incident report the authors filed with California's emergency services office on 13 September, after that office had said the incident did not qualify its threshold, and comparing California's reporting form with EU requirements shows the latter has no field for who was harmed, so the victim may go unrecorded. The third is suggested amendment text for the draft US AI Kill Switch Act, which currently excludes incidents during testing, such as this incident.

    However, the authors state that they have not read the amended RAISE Act or the Commission's incident reporting template in full, so their analysis rests partly on law-firm summaries. The suggested amendment also keeps the testing exemption only if the company can prove that its model never left the test environment, with a government official deciding over the matter; the paper should explain why the burden and the decision sit there, and whether and how the company could mount challenges.

    Read full reviewShow less

Cite this project

@misc{gilly2026one,
  title = {{One Incident, Four Regimes: An Evidence-Sufficient Request for Information, a Filed Critical Safety Incident Report, and Two Fixes to the AI Kill Switch Act}},
  author = {Travis Gilly},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/one-incident-four-regimes-an-evidencesufficient-request-for-information-a-filed-critical-safety-incident-report-and-two-fixes-to-the-ai-kill-switch-act-dm64}},
  url = {https://apartresearch.com/sprints/projects/one-incident-four-regimes-an-evidencesufficient-request-for-information-a-filed-critical-safety-incident-report-and-two-fixes-to-the-ai-kill-switch-act-dm64}
}

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