Skip to content
Apart Research
Apart Research
Contact us
Sprint projectSep 13, 2026Kabirwala, Pakistan

Draft Request for Information under Regulation (EU) 2024/1689: Regulatory Response to Unsanctioned Agent Behaviour During Cyber Testing and the July 2026…

Muhammad Akash Awan, Sarah Jones, Sunil Franklin · Team Article 91

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

Read the report

Report: Draft Request for Information under Regulation (EU) 2024/1689: Regulatory Response to Unsanctioned Agent Behaviour During Cyber Testing and the July 2026…

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

This report presents a draft Request for Information (RFI) prepared under Article 91(1) of Regulation (EU) 2024/1689, the AI Act. The RFI concerns publicly reported unsanctioned agent behaviour during cyber testing and the July 2026 Hugging Face incident. Its purpose is not to determine liability. Rather, it provides a proportionate mechanism for the European AI Office to obtain the documentation needed to assess whether the relevant provider and model fall within the AI Act, whether a general-purpose AI model may present systemic risk, and whether the provider has met the documentation, evaluation, incident-reporting, and cybersecurity obligations that apply. The report translates the draft RFI into a structured research submission. It sets out the legal basis, public-information context, safeguards for confidential and cybersecurity-sensitive material, and the 19 categories of information sought in Annex I. It also explains why the request is organised around verifiable records: the incident timeline, model and configuration status, evaluation conditions, evidence integrity, containment, and mitigation. The result is a regulator-facing instrument designed to support fact-finding without presuming that the public account is complete or that an infringement has occurred.

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. This document collates and concretises what the Office would ask in an initial set of RFIs on the incidents, with direct references to the relevant Articles of the Act. It could be directly useful to the Office, and the specific information it would elicit also looks useful for any subsequent enforcement action.

  2. This submission drafts an Article 91(1) Request for Information for the July 2026 Hugging Face incident. Question 3 is particularly strong – instead of leaving scope to the provider, it considers whether the Article 2 research exemption survives when a model reaches Union infrastructure through its own unauthorised action rather than deliberate deployment. The evidence design is good, particularly the timestamp sequence separating occurrence, technical detectability, automated detection, human review and containment, and the requirement for corroborating records held outside any agent-accessible environment.

    Question 12 departs from the submission's posture by asking for findings that "concluded the zero-day had been patched" and why the agents' recreation of the message board went undetected, which could be reframed as a more open request without presupposition to keep the posture consistent. One question runs to eleven sub-categories of controls, another seeks every attempted interaction whether successful or not, with no narrowing rationale: a shorter first-instance request with a defined escalation path would be better.

    Read full reviewShow less

Cite this project

@misc{awan2026draft,
  title = {{Draft Request for Information under Regulation (EU) 2024/1689: Regulatory Response to Unsanctioned Agent Behaviour During Cyber Testing and the July 2026 Hugging Face Incident}},
  author = {Muhammad Akash Awan and Sarah Jones and Sunil Franklin},
  year = {2026},
  month = sep,
  note = {Submitted to AI Incident Response Sprint, an Apart Research Sprint},
  howpublished = {\url{https://apartresearch.com/sprints/projects/draft-request-for-information-under-regulation-eu-20241689-regulatory-response-to-unsanctioned-agent-behaviour-during-cyber-testing-and-the-july-2026-hugging-face-incident-851g}},
  url = {https://apartresearch.com/sprints/projects/draft-request-for-information-under-regulation-eu-20241689-regulatory-response-to-unsanctioned-agent-behaviour-during-cyber-testing-and-the-july-2026-hugging-face-incident-851g}
}

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