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

Strengthening International AI Incident Discovery

İpek Cerrahoğlu, Ömer Bilgin, Aydin Yigit Alfatli · Team A-O-I

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

Read the report

Report: Strengthening International AI Incident Discovery

Open in a new tab (opens in new tab)
More on drive.google.com (opens in new tab)
Share

Mandatory AI incident reporting exists to make dangerous events visible to regulators while there is still time to act. We coded the six such regimes in force or set to be enforced soon from primary legal texts and applied each trigger to the documented loss of control incidents of summer 2026, finding that none clearly compel a report. Disaster-scale thresholds are only part of the defect: most triggers require harm before engaging, almost every text exempts evaluation findings, and no binding instrument requires near misses to be reported. We set out recommendations for four stakeholder groups, mostly needing no new legislation: a reporting ladder whose incident tiers capture all real-world harm below the crisis threshold, measurable near-miss triggers with denominators developers can adopt unilaterally, and information-sharing channels between jurisdictions.

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. This is interesting analysis that supports the claim that most people agree with -- we're not doing a good job defining serious incidents and when they should be reported. I particularly appreciate recommendation 6.3.1 (Fund the external detection layer as reporting infrastructure) and would recommend follow up with grant organizations or "applied" fellowships to explore what this could look like in the real world.

  2. This project conducts an analysis of whether recent control incidents would cross the threshold for legally required incident reporting under six regulatory regimes (South Korea, Vietnam, EU, California, Illinois, New York). The paper concludes that only the EU AI Act would have (likely) required reporting these incidents. The paper then accurately identifies some flaws in existing incident reporting regimes that result in these incidents failing to trigger the reporting thresholds.

    Strengths:

    • Presentation: Paper is well written and easy to follow.
    • Importance: Paper addresses an important problem in AI incident reporting.
    • Practicality: Paper produces useful analysis and recommendations.
    • Soundness (results/discussion sections): Analysis appears to be mostly sound, and authors cite most relevant literature.

    Weaknesses:

    - Soundness (recommendations section): Recommendations are not sufficiently justified or analyzed in sufficient depth.

    For instance:

    (1) In 6.1.1, the trigger scope should plausibly be simply independent of deployment status and use case, rather than being pegged to these factors at all;

    (2) In 6.1.2, presumably a tiered reporting system would have different reporting obligations at each tier, but this is not discussed in the paper;

    (3) In 6.1.3, additional incident detection/reporting within jurisdictions is needed to enable cross-border information sharing (as I think is recognized in other sections of the paper but not here), and moreover, it's not obvious that cross-border incident sharing would be particularly helpful (e.g., if most of the targets that an LLM would consider "high value" are in the US, then we may not really be learning much from anything in other jurisdictions).

    (4) In 6.2, many of the recommendations seem rather generic / difficult to operationalize (e.g., "improve internal evals and sandboxing")

    Read full reviewShow less

Cite this project

@misc{cerrahoglu2026strengthening,
  title = {{Strengthening International AI Incident Discovery}},
  author = {İpek Cerrahoğlu and Ömer Bilgin and Aydin Yigit Alfatli},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/strengthening-international-ai-incident-discovery-d0t9}},
  url = {https://apartresearch.com/sprints/projects/strengthening-international-ai-incident-discovery-d0t9}
}

More from AI Incident Response Sprint

All 231 entriesBrowse all projects

Build something like this at the next Sprint

AI Collusion Research Sprint · Oct 23 - 25, 2026