Observed 9 September 2026. Bounded primary-source return; no canonical edits, outreach, payments or account actions. The revealing record is the campaign's own acknowledgements: it attributes drafting, text feedback, management and announcement brokerage to named people. These establish work performed, not final approval rights, donor control or government adoption.
S1 — Specific formation roles
Source: [campaign acknowledgements](http[local research file] undated; complete substantive page read, reader lines 2–24. Retrospective organizer attribution, access_state read. The February and July references have no year; interpreting them as 2025 follows the dated launch, rather than an explicit year here. Short extract: “drafted the first versions of the call in February”.
Candidate directional edges from this source:
| Actor | Supported action → object | Locator |
|---|---|---|
| Charbel-Raphael Segerie | Researched and drafted → initial call; coordinated stakeholders until July TFS partnership | Initiation |
| Florent Berthet; Manuel Bimich | Discussed wording → statement | Statement text |
| ControlAI; FLI; SaferAI; Marc Rotenberg | Provided feedback → statement | Statement beta testing |
| Arthur Grimonpont | Drafted → initial FAQ; pitched → Maria Ressa | FAQ; UN Announcement |
| Niki Iliadis; Su Cizem; Tereza Zoumpalova | Significantly revised → FAQ | FAQ |
| Niki Iliadis; Tereza Zoumpalova | Managed → project within TFS | Management |
| Florent Berthet; Manuel Bimich | Identified → key signatories | Signatory collection |
| Niki Iliadis; Anoush Tatevossian | Conducted policy/CSO outreach → signatory collection | Signatory collection |
| Jared Perlo | Suggested → Segerie's TFS collaboration | Advisors |
| Anoush Rima Tatevossian | Advised → UN engagement | Advisors |
These are attributed contribution edges, not contracts or delegated approval powers.
S2 — A concrete choice upstream of the campaign
Source: [Ten AI Governance Priorities: Survey of 44 Civil Society Organizations](http[local research file] The Future Society, 2 April 2025. Access_state read; substantive methodology, results and authorship sections read, especially reader lines 79–109 and 187–190.
Its methodology reports that a Paris AI Action Summit side event on 11 February 2025 generated 15 priorities from 61 participants. Renaissance Numérique and TFS consolidated those into 10 voting options, then surveyed 44 anonymous, self-identified organizational representatives on 15–25 February. Red lines ranked first at 7.25/10 under a weighted ranking method. Authors: Caroline Jeanmaire, Jessica Galissaire, Tereza Zoumpalova and Niki Iliadis. The authors explicitly limit representativeness.
Candidate edges: Renaissance Numérique + TFS → consolidated voting options; respondent group → ranked the offered red-lines option highest; TFS publication → supplied a survey subsequently linked by the campaign FAQ. This establishes agenda curation and a cited justification. It does not show an unprompted population-wide consensus, identify every respondent, or establish that the survey caused the first draft. The actual 15-to-10 mapping and original questionnaire would discriminate what was merged or excluded.
S3 — Dated launch and initial public framing
Source: [22 September 2025 campaign press release](http[local research file] The current site also links the [Foreview-hosted copy](http[local research file] Access_state read; all three PDF pages read. These are the same release family, not two independent confirmations; original byte equality was not tested.
PDF page 1 identifies CeSIA, TFS and Berkeley CHAI as co-organizers alongside 20 partners, and reports over 200 individuals and 70 organizations. Page 3 says the call was “announced this morning by Maria Ressa” at the high-level meeting marking the UN's 80th anniversary, and schedules presentation at AI Safety Connect at UN Headquarters for 25 September. The latter is a planned event in this dated document; the announcement is an organizer-reported completed action. Neither establishes UN endorsement. The release seeks an enforced international agreement by end-2026; it does not present a negotiated treaty text.
Candidate edges: CeSIA/TFS/CHAI → co-organized call; Maria Ressa → announced call on 22 September 2025; campaign → scheduled presentation at AI Safety Connect on 25 September. The government-uptake lane owns independent UN records.
S4 — Documented route into summit programming
Source: [Athens Roundtable agenda](http[local research file] nine-page program for 4 December 2025, London. Access_state read for PDF pages 1–6; remaining pages only encountered as returned context. Its filename labels it V0; no assertion that this was the last issued agenda.
Page 1 names TFS as organizer with OECD and UNESCO and calls the event an official pre-summit event of the 2026 AI Impact Summit. Pages 4–6 explicitly build Dialogue 1 on the Global Call and a pre-event survey; participants would be grouped using survey responses and backgrounds, then produce a roadmap for later diplomatic forums. This is a designed agenda pathway, not just shared attendance.
Candidate edge: TFS-organized Athens Dialogue 1 → used call and survey to structure agenda. The [TFS outcome article dated 12 February 2026](http[local research file] reports the dialogue happened and informed its report. Its substantive body was read; the linked report PDF returned a reader error and was not read. Agenda design and organizer-reported execution remain separate evidence states.
Source: [India workshop findings](http[local research file] three pages, dated 18 February 2026; all read. Page 1 names TFS and CeSIA as conveners at AI Safety Connect Day during the India AI Impact Summit, and expressly links the session to the call. Pages 2–3 record unresolved disagreements over global harmonization, standards and balancing immediate versus frontier harms. Candidate edge: TFS + CeSIA → convened call-linked summit workshop. This report does not establish that governments adopted the call or agreed a single red-line regime. The March URL folder is not treated as the event date.
S5 — Distinct from A Narrow Path
[A Narrow Path's own changelog](http[local research file] substantive entries read, dates its public V1.0.0 release to 2 October 2024, V1.1.0 to 19 December 2024, and V1.2.0 to 3 April 2025. Its [main page](http[local research file] and [About page](http[local research file] portions read, identify Andrea Miotti, Tolga Bilge, Dave Kasten and James Newport as authors and describe a three-phase strategy, beginning with preventing superintelligence development for 20 years. This is a different document and project; its named authorship and dated release cannot be transferred to the September 2025 call.
The shared network does matter: S1 specifically credits ControlAI feedback, while the current [call and FAQ](http[local research file] read in the statement, organizer, submission and FAQ sections, seek international agreement but leave particular red lines for scientific-diplomatic definition. The FAQ expressly says it need not reflect every signatory's views. Endorsing the call therefore does not establish endorsement of every FAQ example or of A Narrow Path's full program.
Remaining decision frontier
The inspected site reviews signature submissions before publication, but that does not identify who selected partner organizations or who could veto wording. No founding agreement, tracked draft, editorial approval record, partner invitation criteria or decision-rights instrument was obtained. Nor was a complete version history of the call or FAQ found: S1 describes early versions, and S3 anchors public launch, but present FAQ wording must not automatically be backdated to that launch.
The strongest rival to a centrally controlled private-policy project remains an openly assembled coalition whose members provided different contributions without accepting a common detailed platform. The next discriminating records are precise: the February draft and revision history; the July CeSIA–TFS partnership record; the final approval record for the September statement and later FAQ; the partner-selection record; and the survey's option-consolidation mapping. Custodians are CeSIA/TFS and their named project managers, with Renaissance Numérique for the survey. Pursue only if publicly released or newly linked; no outreach is authorized.
Acquisition stops at this coherent evidence breakpoint. Public-reader source text and PDF scopes were inspected; this Markdown is the only local artifact created by this lane, so no original-file hashes or preservation claims are made. The broad Athens recap PDF also returned a reader error; neither failed PDF is represented as read. Search-result-only material, including the Manifund funding appeal, was passed to the appropriate lane and is not evidence for this return.