Aller aux articles
AI BroadsheetTHE BRIEF
LA SÉLECTION
01 / 14

18 oct. 2026

Anthropic réduit l'embauche de débutants alors que les agents d'IA effectuent plus de tâches

La part des postes subalternes est passée de 27 à 9 % des nouvelles embauches chez Anthropic, où les agents d'IA enregistrent six fois plus d'heures.

Rédaction assistée par IA
02 / LE CONTEXTE

Ce qui change ?

Ce recul des embauches laisse moins de débouchés aux chercheurs d'emploi débutants dans le secteur technique, les entreprises s'en remettant aux agents d'IA pour exécuter les tâches de base.

À RETENIR

Consultez l’article complet pour les sources, les limites et les mises à jour.

Lire l’articleUne mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

Toutes les sources de l’article

Rédaction assistée par IA

Glisser pour le contexte
02 / 14

22 oct. 2026

Mind & Society: “I trained the AI that took my job”

A Malta designer says their team built an AI, then lost their jobs to it. Goldman Sachs now sees 9% of US workers displaced within a decade, and beginners feel it first. What professionals can do.

Captions, no narration · credits in sources
02 / LE CONTEXTE

Ce qui change ?

A designer in Malta says: I helped train the AI that took my job. Speaking to the Times of Malta this month, the designer said their team built an AI to make design decisions. Managers then judged it could replace half the team, the designer says. The employer wasn't named, and the industry body disputes reports of big AI job losses. Goldman Sachs now estimates AI could displace about 9 percent of US workers within a decade: some 15 million people. The first jobs going are in tech, consulting and graphic design. And beginners feel it first: Stanford found 16 percent fewer jobs for 22 to 25 year olds in the most exposed fields. Experts disagree on how much is really AI. Andrew Challenger, who tracks layoffs, says some jobs are literally being replaced. The research is clear on one thing: jobs shrink where AI does the whole task, and hold where it helps a person. Own the judgment and the…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

IBTimes UK (citing Times of Malta, 18 Oct 2026)Allwork.Space (Goldman Sachs estimate)Stanford Digital Economy Lab: Canaries in the Coal MineCrunchbase News

Captions, no narration · credits in sources

Glisser pour le contexte
03 / 14

POUR COMPRENDRE L’IA

Le score ne raconte pas toute l’histoire.

Un bon résultat à un test d’IA ne garantit pas la même performance dans la vie réelle.

Illustration IA · texte assisté par IA
02 / LE CONTEXTE

Ce qui change ?

Le NIST distingue le résultat sur un test fixe de la performance estimée sur un ensemble plus large de tâches similaires. Les hypothèses et l’incertitude comptent.

À RETENIR

Demandez ce qui a été testé, par qui et dans quelles conditions. Cette explication ne classe aucun modèle.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

NIST · AI 800-3 · February 2026

Illustration IA · texte assisté par IA

Glisser pour le contexte
04 / 14

POUR COMPRENDRE L’IA

Ils cherchent les failles. Exprès.

Une équipe rouge pousse un système vers l’échec pour découvrir ses points faibles.

Illustration IA · texte assisté par IA
02 / LE CONTEXTE

Ce qui change ?

Selon le NIST, cette démarche structurée emploie souvent des méthodes adverses. L’IMDA souligne le rôle des langues, des cultures et de l’expertise des testeurs.

À RETENIR

Ne pas trouver de faille pendant un exercice ne prouve pas qu’il n’en existe aucune. Testez uniquement des systèmes pour lesquels vous avez une autorisation.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

NIST · Red teamingSingapore IMDA · January 2025

Illustration IA · texte assisté par IA

Glisser pour le contexte
05 / 14

POUR COMPRENDRE L’IA

Un grand score. Une petite question.

Avant de croire un classement d’IA, demandez : que mesure-t-il vraiment ?

Illustration IA · montage original
02 / LE CONTEXTE

Ce qui change ?

Un test contrôlé fournit des indices utiles. Les outils, les questions et le contexte de déploiement peuvent changer les résultats.

À RETENIR

Vidéo muette avec texte à l’écran. Aucune garantie de performance dans le monde réel.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

NIST · AI 800-3 · February 2026

Illustration IA · montage original

Glisser pour le contexte
06 / 14

17 oct. 2026

Meta retire une clause sur l'aspiration de contenus de presse pour entraîner son IA

L'entreprise a révisé ses conditions après des critiques visant des termes qui laissaient entendre qu'elle pouvait collecter les contenus d'éditeurs en ligne.

Rédaction assistée par IA
02 / LE CONTEXTE

Ce qui change ?

Le changement modifie les règles écrites déterminant si les reportages des sites d’information et des éditeurs en ligne peuvent être moissonnés pour l’intelligence artificielle. Pour les créateurs, cette mise au point répond aux tensions entourant la collecte de contenus numériques sans autorisation explicite par les géants technologiques.

À RETENIR

Consultez l’article complet pour les sources, les limites et les mises à jour.

Lire l’articleUne mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

Toutes les sources de l’article

Rédaction assistée par IA

Glisser pour le contexte
07 / 14

11 oct. 2026

Satya Nadella de Microsoft réclame un frein d'urgence pour l'IA

Satya Nadella affirme que les modèles d'intelligence artificielle doivent être considérés comme compromis et nécessitent un frein d'urgence.

Rédaction assistée par IA
02 / LE CONTEXTE

Ce qui change ?

Ces déclarations incitent les organisations et les usagers à cesser de présumer que les systèmes automatisés avancés sont sûrs par défaut.

À RETENIR

Consultez l’article complet pour les sources, les limites et les mises à jour.

Lire l’articleUne mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

Toutes les sources de l’article

Rédaction assistée par IA

Glisser pour le contexte
08 / 14

21 oct. 2026

The Sir Jack Files, special edition: The Panic That Wasn't

On October 30, 1938, Orson Welles put Martians on the radio, and by morning America had a legend: mass panic. The evidence says it mostly didn't happen. A three-and-a-half-minute film on how newspapers built…

AI visuals & voice
02 / LE CONTEXTE

Ce qui change ?

New York. Sunday, October thirtieth, nineteen thirty-eight. Eight o'clock. On CBS radio, a twenty-three-year-old named Orson Welles begins a Halloween play. It's The War of the Worlds. Martians land on a farm in Grover's Mill, New Jersey. And the story is told as fake news bulletins, breaking into a dance band. By the next morning, America had a legend. It has been told ever since. The trouble is, it's mostly not true. Monday's New York Times, front page: radio listeners in panic, taking war drama as fact. The papers told of families fleeing their homes, of switchboards jammed with calls. The next day, unshaven, Welles faced the press. A reporter asks: Reporter: “Were you aware of the terror going on throughout the nation while you were giving the broadcast?” Welles: “Oh no, of course not. I was frankly terribly shocked to learn it did.” Another: Reporter: “Should you have toned down…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

Jefferson Pooley & Michael J. Socolow, “The Myth of the War of the Worlds Panic,” Slate (Oct. 28, 2013)NPR (Oct. 30, 2013)Radio Guide transcript of Welles's press conference, Oct. 31, 1938 (via Wellesnet)A. Brad Schwartz, Broadcast Hysteria (Hill and Wang, 2015); letters at the University of MichiganW. Joseph Campbell, Getting It Wrong (University of California Press, 2010)Hadley Cantril, The Invasion from Mars (Princeton University Press, 1940)Radiolab, “Could It Happen Again?” (2010), on Quito, 1949College of Wooster archive: fire at Radio Quito, 1949

AI visuals & voice

Glisser pour le contexte
09 / 14

21 oct. 2026

Op-Doc: The Good Room

Millions of people have visited Lascaux. Almost none of them have been to Lascaux. A four-minute opinion film on copies better than the original, AI answers that arrive too complete, and the discipline of…

AI voice · credits in sources
02 / LE CONTEXTE

Ce qui change ?

In September nineteen forty, four teenagers climbed down a hole in the ground in south-western France, and found a room with animals on the walls. Horses. Stags. Aurochs. A bear. Painted with charcoal and ochre, around twenty thousand years ago. In nineteen forty-eight the cave opened to the public, and it did what any wonder does. It drew crowds. At its peak, about eighteen hundred people a day. Within a few years, the paintings began to sicken. Not from vandalism. From breath, and heat, and electric light. Green algae. White crystals. Looking at the paintings was destroying the paintings. So in nineteen sixty-three, France closed the cave. And built another one. The newest copy opened in December twenty sixteen, built from millimetre-scale scans. It is better lit than the original ever was. There are no queues that damage anything. You can stand in front of one horse for as long as…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

French Ministry of Culture, LascauxPew Research Center (July 2025)

AI voice · credits in sources

Glisser pour le contexte
10 / 14

20 oct. 2026

The Sir Jack Files #11: The Case of the Ghost Shopper

Stripe says AI agents' purchases through Link grew 38-fold in September, but won't say how much money that is. Who's shopping, what they buy, and how to keep the final yes for yourself.

AI voice · credits in sources
02 / LE CONTEXTE

Ce qui change ?

Case eleven. Someone is shopping with your money. And it isn't you. It's your AI. September. Stripe says purchases made by AI agents through Link, its payment service, grew thirty-eight times in one month. Payment volume: up twenty times. How much money is that? Stripe won't say. And when you start small, big multipliers can hide small numbers. That's clue number one. Clue two: who's shopping. About seventy agents when Stripe looked. Around two hundred now, according to Stripe executive Dan Hill, including Meta's Muse. Most spend about a hundred dollars a purchase. Some book flights, or buy furniture worth thousands. Retail leads. Travel is next. The safety catch: the agent never sees your real card. Link hands it a separate sixteen-digit number, and you can set a limit. So before you hand an assistant your wallet: set the limit. Read every receipt. And keep the final yes for yourself.…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

The Decoder, Maximilian Schreiner (Oct 19, 2026)Upstarts Media (cited by The Decoder)

AI voice · credits in sources

Glisser pour le contexte
11 / 14

20 oct. 2026

The Sir Jack Files, special edition: The Wrong Face

In 2020 a Detroit father was arrested on his driveway, in front of his daughters, because facial recognition matched his old licence photo to a blurry shoplifting video. A four-minute motion comic from court…

AI visuals & voice
02 / LE CONTEXTE

Ce qui change ?

Detroit. January ninth, twenty twenty. Twenty-two minutes past five. A man pulls into his own driveway after a day at work. His wife is at the door. His two little girls are watching. Seconds later, a police car pulls in behind him. Handcuffs. He asks what for. The warrant says: larceny. His wife asks for a card, some way to reach him. An officer tells her: Officer: “Google it.” His name is Robert Williams. And the witness against him was a machine. Rewind. October second, twenty eighteen. A man walks into a Shinola store in downtown Detroit, and walks out with five watches. About thirty-eight hundred dollars. The cameras catch him, barely. Head down. A red St. Louis Cardinals cap hiding his face. March twenty nineteen. A detective takes one blurry frame from that video and sends it for a facial recognition search. Michigan State Police run it. The software, from a company called…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

Williams v. City of Detroit, First Amended Complaint (E.D. Mich., 2023)NPR (June 24, 2020)ACLUAP (June 2020; June 2024)Michigan Public (June 28, 2024)NBC News (Porcha Woodruff, 2023)NIST, NISTIR 8280 (2019)Georgetown Law, The Perpetual Line-Up (2016)

AI visuals & voice

Glisser pour le contexte
12 / 14

20 oct. 2026

The Sir Jack Files #10: The Case of the Missing Junior

Anthropic studied its own staff. Junior technical hires fell from 27% to 9%, and its AI agents now log six hours for every human hour. Sir Jack follows the empty chair.

AI voice · credits in sources
02 / LE CONTEXTE

Ce qui change ?

Case ten. In case one, I went looking for a body, and found none. This time, there's an empty chair. October seventeenth. Anthropic, the company behind Claude, publishes a report on its own workforce. A lab measuring what its machine did to its own office. Early twenty twenty-five: juniors were twenty-seven percent of its technical hires. Now: nine. Since March, its AI agents have logged more hours than its people. By August: six agent hours for every human hour. And the humans? More meetings. Up twenty percent in a year. One in five staff said their job had shifted from doing the work to managing Claudes. The report admits Anthropic is a special case. And more code does not necessarily mean better code. But here's the clue nobody can ignore. Every senior was once a junior. Cut the first rung, and who's left, in ten years, to check the machine? If you're starting out: learn to judge the…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

TechCrunch, Anthony Ha (Oct 17, 2026)The Decoder (Oct 18, 2026)Anthropic, Employment at Anthropic

AI voice · credits in sources

Glisser pour le contexte
13 / 14

20 oct. 2026

The Sir Jack Files #9: The Case of the Burglar Who Never Slept

One hacker, a few lines of instructions and an army of AI agents: more than 1,200 shopping sites tried, more than 100 companies hit, including John Deere and SharkNinja. What Reuters found, and the basics that…

AI voice · credits in sources
02 / LE CONTEXTE

Ce qui change ?

Case nine. A burglar who never sleeps, never gets bored, and works for about fifty dollars a job. September twenty-fifth. One hacker types a few orders. Fewer than a hundred Chinese characters. Then lets the machines go to work. AI agents, running on an open-source platform called Cairn, start trying doors. Shopping sites in the US, Canada, Australia and Japan. More than twelve hundred of them. A typical break-in: about twenty-three hours of patient, automated probing. When the front door held, the agents went round the back. A forgotten control panel. A flaw. Hours of trial and error. More than a hundred companies were hit. Reuters named some: John Deere and SharkNinja. Deere says there's no sign sensitive information was accessed. SharkNinja says the site held only product support, and the threat is contained. The researcher who found it says it's not script kiddie stuff. His…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

Reuters, Raphael Satter (Oct 19, 2026)Quartz

AI voice · credits in sources

Glisser pour le contexte
14 / 14

19 oct. 2026

The Sir Jack Files #8, Part 2: The Case of the Ship That Never Sank

Jack and Rose were fiction, but a grave marked J. Dawson is real. Why a 1997 film and a 1912 ship still hold millions: the people on board.

AI voice · credits in sources
02 / LE CONTEXTE

Ce qui change ?

Case eight, part two. Nineteen ninety-seven. A love story on a sinking ship becomes one of the most watched films in history. Titanic took in more than two billion dollars and won eleven Oscars. For twelve years, no film earned more. Was Jack real? No. Jack and Rose were invented. But in Halifax, a grave marked J. Dawson belongs to Joseph Dawson, a coal trimmer who died that night. Fans still visit. The director says the name was a coincidence. Here is the secret. The film turned a statistic into faces. About sixty-two percent of first-class passengers survived. In third class, about one in four. In twenty twenty-two, deep-sea robots took about seven hundred thousand images of the wreck and built a full digital twin. Technology lets us see the ship. A story lets us feel it. Even now, new Titanic videos, some made with AI, find millions of viewers. Because Titanic never sank from memory.…

À RETENIR

This is an AI-assisted explainer. Check the linked reporting for the original evidence.

Une mise au point par jour.
03 / LES SOURCES

Curieux ?
Vérifiez.

Voici d’où viennent les informations.

CBC: J. Dawson grave, HalifaxEncyclopedia Titanica (Joseph Dawson; survival by class)PetaPixel: first full-sized Titanic scans (2023)

AI voice · credits in sources

Glisser pour le contexte

VOUS AVEZ FAIT LE TOUR

Un peu plus de recul.

C’est toute cette édition. Soufflez, explorez un thème ou ouvrez le journal.

Ouvrir le journal