What Must the Numéro 0 Prove Before GISI’s Plan Can Proceed?
GISI’s plan would eliminate all eight of its secrétaire-de-rédaction posts. A French court has suspended it, in the order’s own terms, until real‑condition tests and numéros 0 confirm which of those tasks its AI tool can take.
On August 4, 2026, someone sat down in front of a bailiff to test the system GISI was building its newsroom around. Three articles went through DIGI, the tool meant to automate the correction work of GISI’s secrétaires de rédaction, the SRs, GISI’s copy desk. The Créteil court’s order records who ran the test: GISI’s own AI-project lead, well acquainted with the tool but not shown to hold the status of rédacteur. No one from the other side was present; the order records the test ran “non-contradictoire,” one-sided by design. What came back ran the wrong direction. The “optimized” drafts came out longer than the originals, leaving the rédacteur to make “recoupes,” cuts.
That test sat in the record of a plan the Créteil court had already ordered suspended: a reorganization that would eliminate all eight of GISI’s secrétaire-de-rédaction posts and create two chef-d’édition posts to absorb what DIGI cannot do alone. GISI contends, in its own defense, that DIGI can carry about 70 percent of an SR’s working time, leaving 30 percent for a human editor — its own workload analysis, pleaded as fact, not a finding the court reached.
The company’s own arithmetic
GISI’s own arithmetic did not agree with itself. The order records that GISI’s internal documents put the residual SR workload at 30 percent in one estimate, then at 25 to 30 percent in another. The same task, two different numbers, from the same company. The CSE’s outside reviewer was the expert firm ADDHOC. GISI’s own evaluation, the report recites, put the residual workload at 2 to 2.4 full-time-equivalent posts: two to two-point-four full jobs, not a fraction of one.
Against its own 2-to-2.4 FTE estimate, GISI chose to recruit for only two chef-d’édition posts. The expert’s verdict on that math, the court recites: “fortement insuffisant,” strongly insufficient, to carry out the editing and coordination work.
The same report forecasts a risk of major intensification of work for rédacteurs, the order records. It ties that risk to the proofreading GISI’s plan would remove, and to chef-d’édition posts the expert had already called under-sized. On our reading, that is the shape of the story GISI’s plan was set to tell: a job title it would eliminate, and arithmetic that was never asked to agree with the staffing meant to replace it. A court has now suspended that story until the numbers do.
What the plan didn’t count
The CSE, GISI’s works council, alleges the plan does something simpler than eliminate a job: it moves the SR’s correction work onto the rédacteur, unmeasured, as a new task the CSE calls “post-édition.” GISI’s own May 2026 account had called the changes to the rédacteur’s role minor: in its own words, “quasiment aucun impact,” almost no impact, on workload or hours. Its own 2025 workplace risk assessment had already named three AI-related risks to staff (job-insecurity stress, organizational disruption, insufficient AI guidance), and the order records that GISI never scored any of them for frequency or severity.
GISI’s own minutes trace how the timeline slipped. In October 2025, the company told its works council it would need to test AI functionality within the SRs’ own task scope before going further. By May 2026, management was defending the pace of the rollout against staff objections by pointing to the failure of earlier AI working groups, where some participants, the minutes record, had been called “collabos.” That slippage runs through the whole timeline the order reconstructs: a task first scoped in October, an impact assessment revised in May, and a numéro 0 deadline reported to the same works council whose objections GISI was, by its own account, moving fast to overcome. GISI’s own account, by month’s end, was that trial issues had been produced without going as far as the layout. The Créteil court’s order records that GISI had not shown it carried out the complete numéro 0: as of the hearing, GISI had not shown it finished the one it had announced for the end of June.
The court’s answer
The Créteil court ruled that GISI’s reorganization, and the collective-dismissal plan for the eight SR posts that depends on it, is suspended until GISI does the test it has not shown it finished: real-condition trials and a complete numéro 0, run to confirm which SR tasks the AI tool can do and which it cannot. The same order also records a further condition: where warranted, GISI must reassess what its plan leaves on the desks of the employees who stay (rédacteurs, typographes, and maquettistes among them) before the reorganization can go forward.
Non-compliance carries a price, though not the one a headline suggests: an astreinte of €8,000 per day of delay, running up to three months. That is a penalty for delay, not a fine already owed, and well short of the €20,000-a-day the CSE had asked the court to impose. Nor is it GISI’s first astreinte here: a July 2025 référé had already suspended GISI’s AI rollout pending CSE consultation, under a smaller €1,000-a-day penalty, confirmed by the Cour d’appel de Paris on May 21, 2026, with a further €3,000 award.
The court also found that neither GISI’s liability nor the amount claimed had been sufficiently established, and declined a damages award beyond the suspension itself. GISI, the order records, employs 127 people, including 92 journalists by the court’s count (91, GISI says, by its own).
Who said so, on the record
Pablo Aiquel — a journalist at La Gazette des communes and an SNJ-CGT union delegate, per L’Informé — called the ruling “une décision historique qui souligne les manquements importants de notre direction,” a historic decision underscoring serious failures by his own management, he told the outlet. Aiquel, identified separately by the European Federation of Journalists as General Secretary of the SNJ-CGT and Vice-President of the EFJ, put the stakes more broadly: “This first legal battle victory is a signal to journalists, in France and in Europe, who would be threatened with being replaced by artificial intelligence agents: fight, don’t let it happen!”
Infopro Digital’s management, contacted by L’Informé and referred to in its reporting only as “la direction,” said it would comply: “nous avions prévu de poursuivre les tests déjà en cours depuis plusieurs mois. Comme le juge nous le demande, nous suspendons le projet le temps de finaliser ces tests” — it had already planned to continue tests under way for several months, management said, and would suspend the project, as the judge asked, to finish them.
The Agentics read
Agentics reads this through Digital Derangement Syndrome™ (DDS™): AI-mediated systems failing to recognize legitimate authority. The characteristic that fits here, hedged, is Decision Exclusion: DDS™’s fifth clinical characteristic. Authority Misclassification, the first, runs alongside it. The excluded party in the textbook version of Decision Exclusion is a single credentialed expert. Here it is a job category, and the record above is what that exclusion looks like once someone finally reads it back.
Decision Exclusion, hedged. A decision can exclude evidence that already exists, produced by its own process, simply by never weighing it before shipping. That is the shape of the pattern above, once generalized: a category of work, not a named person, sitting outside the account a plan gives of itself.
Authority Misclassification, contributing. The first clinical characteristic names a failure to associate real authority with its true standing. On our reading, the same failure scales to a job title: work tallied correctly in one document, zeroed out on another.
The correction is procedural, not moral. Nothing here required bad faith, only treating a commissioned answer as a formality, not an input. On our reading, that is the harder discipline: building the decision so paid-for evidence has nowhere left to go but into it.
Where the argument stops
Two people are central to GISI’s own account of its testing, and neither is named in the court’s own order: the AI-project lead who ran the August 2026 bailiff test, and the outside consultant who ran the earlier pilot tests referenced in the May 2026 minutes. We do not name them here either, and nothing above should be read as narrowing who they are. GISI’s 70/30 workload split and the CSE’s “post-édition” characterization are each a party’s own pleaded position, not a finding either side has established on this record. Translations from the French are the editor’s own. No party beyond the cited record was contacted independently for this edition; Infopro Digital’s reply reached the record through L’Informé’s own reporting, not ours. The mapping to DDS™ above is our own editorial reading, labeled as ours, not a finding of the Créteil court.
The record. Tribunal judiciaire de Créteil, ordonnance de référé, 15 September 2026 (Comité social économique de la société GISI c/ S.A.S. GISI), read in full — the order’s own recitation of GISI’s minutes, the ADDHOC expert report, and its operative clause.
The coverage. L’Informé, “LSA, l’Usine Nouvelle… la justice retoque le remplacement des journalistes par l’IA” (Emmanuel Paquette), 15/09/2026, updated 16/09/2026 — Pablo Aiquel’s and Infopro Digital management’s on-record quotes. European Federation of Journalists, “France: First legal battle won for journalists of the Infopro Digital press group threatened with dismissal in favor of AI,” 17‑09‑2026 — Aiquel’s EFJ-quoted statement. ActuIA’s corroborating coverage of the same figures and findings, 17/09/2026, is noted but not separately quoted.