← Red Threads

France/G7 follow-through: CeSIA's proposed diplomatic sprint versus public outcomes

Observed 2026-09-09. Bounded acquisition note for parent review; no canonical case changes. Source identifiers below are local candidate labels, not existing-case IDs. Reuse canonical CeSIA, France/G7, Charbel-Raphaël Ségerie and other established identities during integration.

Finding

There is a documented connection through an event programme and co-organizer relationship, plus a partial functional overlap with the G7 programme. The inspected outcomes do not establish CeSIA's proposed G7 commitment to prepare a coordinated slowdown, and no public transmission or drafting record attributes the relevant official language to CeSIA. That is a bounded result about these records, not proof of no private influence or a verdict on the campaign's whole success.

The strongest new government-side record is the Digital and Technology Ministers' declaration from their 29 May 2026 Paris meeting. Its risk-assessment discussion commitment is more specific than generic AI-safety rhetoric. The strongest new access-side record is a 10 March 2026 Sciences Po/CeSIA programme naming Alexandre Mirlesse, a French foreign-ministry AI/tech diplomat, as a moderator. Neither record closes the connection between campaign proposal and negotiated wording.

Proposed objective and observed outcome

S1 — CeSIA's original public fundraising proposal. The Manifund page, created 4 December 2025, identifies France as its preferred champion state. Its June target is a G7 statement preparing multilateral capacity to slow AI development together following extreme warning events or crossed red lines; it expressly distinguishes preparation from immediate action. Its separate technical track seeks common risk thresholds. It also targets a September 2026 UNGA resolution and claims prior high-level French briefings without naming recipients, dates or documents. These are project aims and self-reported experience. The page's fundraising status does not establish the campaign's total receipts or execution. Locator: “Why CeSIA,” “Track 1,” “Track 2,” June and September calendar items, and “Policy Influence.” [Original proposal](http[local research file]

S2 — Actual ministerial declaration. The five-page French original says ministers met on 29 May. PDF page 2 commits to French-presidency discussions involving the international AI measurement/evaluation/science network and OECD to improve mutual understanding and comparability of AI risk-assessment frameworks, alongside innovation and adoption. This overlaps functionally with harmonization, but does not adopt a common threshold or a slowdown trigger. The complete extracted text has no matches for ralent, seuil, ligne rouge, lignes rouges, CeSIA, red line, moratoire or suspend; the conclusion also rests on reading, not keyword absence alone. Page 1 expressly continues the Hiroshima process dating to 2023 and its later reporting work. Pages 3–5 name OSI, OECD and Inria contributions to other specific outputs, but do not credit CeSIA. Those credits demonstrate that attributable contributions can appear in this instrument; omission does not exclude uncredited input. [Ministerial original](http[local research file]

S3 — Presidential account and announced continuation. On 17 June, Macron's official closing transcript describes a prospective platform among democracies for AI-security cooperation and common standards, with a September meeting planned to finalize work. He frames frontier-model regulation partly around authoritarian access and cyber threats. This is the French President's account of discussions and announced follow-up; it is not the text of a collectively adopted slowdown instrument. His list of nine adopted leaders' statements includes the separate minors call. Locator: AI discussion paragraphs beginning “Lastly” through the September meeting, and opening adopted-statements list. [Official transcript, marked check against delivery](http[local research file]

S4 — Earlier government agenda. An Élysée page published 5 February already identifies safe AI, AI adoption, environmental effects and minors' protection as the digital track's four priorities. It names the responsible ministers and the planned 29 May meeting. This predates the March event. It cannot show that CeSIA had no earlier input, but it prevents attributing the priority's origin to that later event. Locator: four digital axes and meeting calendar. [Official priorities](http[local research file]

The named access connection

S5 — Host institutional programme. Sciences Po identifies its School of Management and Impact, Les Mardis de l'IA and CeSIA as co-organizers of the 10 March event. Mirlesse is scheduled to moderate Alexei Grinbaum's fireside discussion and appears in the closing patronage/organizing acknowledgment. The programme identifies him as an AI/tech diplomat at France's foreign ministry. It also schedules Nicolas Miailhe, Lord Tim Clement-Jones, Indian deputy mission chief Eenam Gambhir and Swiss foreign-ministry official Miriam Minder. Charbel Ségerie introduces Yoshua Bengio. These are named programme roles, not independent attendance proof, civil-service instructions, or evidence that a G7 negotiator received a red-lines draft. Locator: organizers, agenda and closing acknowledgment. [Host programme](http[local research file]

S6 — Organizer's retrospective. CeSIA's 25-page conference report recounts the event and says the French presidency awarded it a G7 label (PDF page 22). Page 10 connects CeSIA's red-lines campaign to its French-presidency ambitions. The report describes France as responding to laboratory leaders' calls, but bundles Davos and the later Delhi summit beside an early-February French announcement. That narrative cannot establish that a later Delhi appeal caused the earlier priority. Earlier appeals or earlier advocacy remain possible. The report also overstates the UN scientific panel's mandate as translating consensus into binding commitments; do not inherit that characterization. A label award letter, its criteria, and any government receipt of the report were not acquired. Label recognition would still require a separate basis before being treated as endorsement of the proposed slowdown. Locator: PDF pages 10 and 22; event agenda pages 23–24. [Organizer's original report](http[local research file]

The report is linked from CeSIA's 30 April retrospective by Pauline Charazac and Charbel-Raphaël Ségerie. This date is a publication marker for the retrospective, not necessarily an embedded PDF version date. [Retrospective page](http[local research file]

Candidate relationships for integration

These are separate edges; do not join them into a causal path without the missing transmission record.

Candidate ID Typed relationship Evidence and stage Boundary
fg7-e1 CeSIA → French G7 presidency: targets as champion-state opportunity S1; reported intention / proposed No government acceptance established
fg7-e2 CeSIA → March10 Sciences Po event: co-organizer S5; source record of programme; retrospective S6 reports event held Organization is not state endorsement
fg7-e3 Alexandre Mirlesse → March10 event: scheduled moderator and named in patronage acknowledgment S5; source record / scheduled role; S6 retrospective is organizer account No inferred G7 drafting mandate or agency-wide approval
fg7-e4 French G7 presidency → March10 event: awarded G7 label S6 PDF22; attributed organizer report Award instrument, criteria and exact endorsement scope missing
fg7-e5 G7 digital ministers → risk-framework comparability discussions: public commitment S2 PDF2; source record / announced future work Not a completed workshop, common threshold, or coordinated slowdown
fg7-e6 French President → democracies AI-security platform and September meeting: announces follow-up S3 AI passage; source record of presidential statement / announced Not a separately inspected collective treaty or CeSIA-attributed outcome

Do not create CeSIA → G7 language “drafted,” “secured,” “caused,” or “adopted” edges from these materials. A possible semantic-overlap observation belongs in an analytical note rather than as an influence fact. Mirlesse's exact programme title is contemporaneous; do not silently substitute an independently assumed senior envoy role.

Rival explanations and the decisive remaining record

The official documents support an independent institutional pathway: earlier Hiroshima/GPAI/OECD work, France's already published agenda, and leaders' stated security/adoption concerns. CeSIA may have reinforced that agenda, sought access after it was established, contributed through an uncredited channel, or failed to secure its specific threshold/slowdown ask. Current evidence discriminates against treating the June outcome as fulfillment of that ask; it does not choose among all these influence alternatives.

The most useful missing object is a dated CeSIA submission and its recipient acknowledgment or a negotiator's draft/change record identifying whether risk-comparability or slowdown language was proposed, accepted, softened or rejected. Likely custodians are CeSIA's policy team and the relevant French foreign-ministry/digital-track or presidential G7 negotiating teams. Event attendance, a G7 label, and matching vocabulary cannot substitute for that object. The label award letter is a separate, narrower record about recognition scope.

As of 9 September, September follow-up and the proposal's September UNGA target are still within their stated target month. Absence from this bounded pass is not a finding that either later event or resolution has failed. No public correspondence, meeting minutes or red-lines draft receipt was recovered in the inspected French/G7 sources.

Acquisition and reading scope

All reads were public originals or publisher-owned accounts. No private APIs, access bypass, outreach or payment. Source bodies were inspected through web readers or ordinary public GET. Only this note is saved under this assignment; the hashes below identify response bytes inspected in memory, not locally retained original files.

Source Origin and extent read SHA-256 of inspected HTTP200 bytes where acquired
S1 Original Manifund proposal; substantive proposal, calendar, experience and success-level sections, current metadata Web reader; no byte hash
S2 French government-hosted PDF; all five pages extracted and read across two outputs; whole-text keyword search d60f5a67035807d2a9ba1c53e0542538585c975037ec966c82445af68c835f43
S3 Official English presidential transcript; AI section, statement list and relevant procedural framing; duplicate body not a second source 178b5a8f6c51df2610593d88bbe33cb0d3f3ae184123d1389910c1bfc79f318b
S4 Official priorities page; complete substantive main body ec35904be329b85c96a0654a582844df9825e16e87b63d2145bc199e206d6e7d
S5 Host's event page; complete substantive event description, programme and acknowledgments; navigation excluded 5cf4bfa296cd4865370bfff7381abdbdd85e4346dc39d9f812c4c5e90390fddc
S6 CeSIA PDF; scoped read pages3–4,10,19–24, with pages10/22 rechecked directly; not a complete close reading of every page 3cd31074aef822a3062f00104c3ca2e58d5a24310ccd917cf9128bd891b91895

The [Élysée 16 June digital-priorities summary](http[local research file] was also read and supplied the original ministerial PDF route. Its publication date must not replace the PDF's 29 May meeting date. The separate minors annex was only initially scoped; this note makes no fresh detailed findings about its provisions.

Bounded searches combined CeSIA/G7 with elysee.fr, diplomatie.gouv.fr and economie.gouv.fr and searched the target slowdown concept. They did not return an attributable government drafting/briefing receipt. This is search-scope evidence, not an exhaustive registry audit. The web reader failed on the ministerial PDF; one ordinary GET substitute succeeded. A transcript link-number click failed, and no claim relies on that unresolved link. Closed UN Geneva403, Security Council PDF202 and UNESCO SharePoint401 routes were not retried.