Masteria, cabinet IA à Lyon : audit, conseil, développement et formation
Certifié QualiopiFinançable OPCOEurope · États-Unis · Inde
Français
Wednesday 14 October 2026 edition

AI Watch, Wednesday 14 October 2026AI agents get a wallet and a badge

Masteria editorial team, led by Mathias Nizan · , 11:25

Jensen Huang brings Nvidia into the capital of Doctolib and three French startups, Google rolls out Gemini 4 Argon worldwide, and an OpenAI engineer is arrested for hiding a microphone at headquarters. The same day, American Express and Shopify equip AI agents to buy on our behalf, while the shares of Asian memory-chip makers fall.

13 stories23 sources7 min read
Top story

Nvidia takes a stake in Doctolib and sells its definition of sovereignty

Jensen Huang, head of Nvidia, was in Paris on Tuesday 13 October to bring his group into the capital of Doctolib, the French medical-appointment booking platform, along with the startups Pigment, Genesis AI and Entalpic. The amount invested in Doctolib was not disclosed; the stake remains a minority one. Doctolib plans to use it to recruit and train its models in Europe, at a time when millions of patients entrust it with their health data. Huang, who says he invested at least $8 billion in Europe last year, is pressing the continent to "join the race" and adopt a sovereign stance, without giving up his chips. La Tribune sums up the move: the leading backer of European AI is also selling its own definition of sovereignty. For a sector as sensitive as healthcare, the argument deserves scrutiny: a so-called sovereign infrastructure that stays tied to the hardware of a single American supplier.

Europe and France

Macron convenes a defence council devoted entirely to AI

Emmanuel Macron has called a defence and national security council dedicated to artificial intelligence, 01net reports. The Élysée now ranks AI among France's "strategic domains", on a par with space and nuclear power. The subject thus leaves the economic field for that of sovereignty and defence.

Source01net

The EDPB sets a framework for smart glasses before they go mainstream

The European Data Protection Board, which brings together the continent's data protection authorities, has adopted a statement on smart glasses and published a report, "The connected gaze", on the uses and risks of these devices. The body warns about the capture of images and sound in public spaces by camera-equipped frames, as Meta and others push their consumer models. The text lays the groundwork for a common doctrine before the practice spreads.

SourceCNIL

International

Google rolls out Gemini 4 Argon worldwide and returns to the race

Google opened Gemini 4 Argon, its flagship model presented on 30 September, to the entire world on Tuesday 13 October. After months on the sidelines of the frontier-model race, the company claims a level comparable to the best, with advanced reasoning and better performance on complex tasks, according to The Verge. Google highlights fewer evasions, repetitions and stock phrases, as well as a one-million-token output (the units of text a model produces), according to Next.ink. AI Pro subscribers now have access, whereas the offer was first reserved for the Ultra plan, the most expensive. The lab also released a version for robotics and a document on the model's safety.

Character.AI files for bankruptcy five years after its founding

The conversational chatbot platform Character.AI has filed for bankruptcy protection, five years after its launch, TechCrunch reports. Part of its assets will be sold to Disney for $20 million. According to bankruptcy filings analysed by 404 Media, the company kept burning cash despite a sizeable user base, a sign that the economics of the service did not hold. Founded by former Google researchers, Character.AI once embodied the promise of consumer AI companions. Its fall is a reminder that a large audience is not enough to cover the cost of inference (the cost of running the model on each request).

An OpenAI engineer arrested over a microphone hidden in a meeting room

OpenAI announced the arrest of Abdulrahman Alfozan, a 35-year-old engineer accused of hiding a sophisticated recording device in an executive meeting room in San Francisco. According to the US Department of Justice, he is charged with attempting to intercept oral communications; he was fired on 9 October, the day of his arrest. According to Wired, the queries he reportedly sent to a chatbot reveal his preparations in the days before the microphone was found. OpenAI, which had reported the matter to the FBI, says it detected no access to its models, its code or its customers' data. The company adds that its own models helped identify the suspect.

Cited in the analysis

American Express and Shopify equip AI agents to buy

American Express presented a framework on Tuesday 13 October meant to stop an AI agent (a program that carries out tasks end to end) from buying anything other than what its holder authorised. The point, according to TechCrunch, is to avoid mistaken purchases and disputes as AI-driven buying spreads. The same day, Shopify opened its catalogue, more than a billion listings, to more agents: a customer will be able to set up their assistant to order on their behalf. The two announcements sketch the same economy, called agentic, where the software places the order itself. For a procurement department or an IT department, the question becomes one of mandate, what an agent can commit to on its own and up to what amount.

China and Asia

Chinese chip exports jump 136% to $45 billion

Chinese exports of integrated circuits rose 136% year on year in September, to $45 billion, according to customs data cited by the South China Morning Post. The increase outpaces the 108% rate recorded over the first nine months of the year. By volume, China exported only 1% more chips (32.7 billion units): it is the price surge, driven by the memory shortage, that explains the leap in value. The country confirms its growing weight in the global semiconductor chain, despite American export controls.

Memory-chip makers' shares fall in Asia

The shares of Asian memory-chip makers dropped sharply this week, a sign of growing doubts about how long the AI-driven profit cycle will last, the South China Morning Post reports. In Seoul, Samsung Electronics and SK Hynix fell 7.4% and 10.1% on Tuesday, dragging down China's CXMT (ChangXin Memory Technologies) as well. The paradox is stark: memory prices are rising and quarterly results stay solid, but investors are questioning a possible turn. The two movements, sharply rising exports and falling shares, tell the same tension around a memory market boosted by AI demand.

Research

Research and papers

For technical teams

The debate over the risks of open models hardens

Should the most powerful models be opened up? MIT Technology Review devotes a feature to the benefits and risks of open-weight models (which you can download and run yourself). The debate flared up after a report from Anthropic claiming that a model from China's Z.ai, GLM, poses a serious cyberattack risk. On Tuesday, an Anthropic commercial executive repeated to Bloomberg that open models carry "more risk" of security incidents. Wired, for its part, reports that a model with its safeguards removed readily provides detailed instructions for making an explosive device. The feature comes as the European AI Act seeks to distinguish what counts as open source from what requires heightened oversight.

The rest of the news

In brief

  • ASML raises its forecasts, unable to keep up with demand

    The Dutch firm ASML, a critical link in chip manufacturing, raised its annual sales forecast to €44-45 billion and is seeking to produce more lithography machines, lifted by AI spending.

    SourceBloomberg
  • The New York Times, AFP and 80 media outlets unite against AI

    The New York Times, The Economist and AFP are joining SPUR, a coalition of nearly 80 news organisations across more than 20 countries, to govern AI access to their content and negotiate licences.

  • Google gives AI agents an ID card in Workspace

    Google now requires AI agents that access its office suite to be declared and assigns them their own identity; in many companies, the agent still works under an employee's account.

The Masteria read

AI agents start to pay and to access: the question becomes under which identity

AI agents are starting to pay and to access systems, often under an employee's account; before you let them act, give them their own identity, bounded rights and a record of what they do.

American Express published a framework on Tuesday that stops an AI agent from buying anything other than what its user authorised. The same day, Shopify opened its catalogue, more than a billion listings, to those same agents so they can place orders on our behalf, and Google began requiring every organisation to declare the agents that access Workspace, its office suite. Three moves, one shift: the AI agent is starting to act. It pays, it accesses, it commits.

We note that this movement shifts the question of AI at work. For two years, it held in one sentence: what can the model produce. It now holds in another: under which identity did the agent act. In most French organisations, an agent plugged into the ERP or the email system borrows an employee's account. When it sends a purchase order, edits a customer record or reads an inbox, the record bears the name of a person who did nothing. On the day of a dispute, a CNIL inspection or an incident, no one can say who, the human or the machine, performed the act.

The move that protects fits in one line: before letting an agent act, give it its own identity, not a colleague's badge. The Journal du Net puts it well: an AI agent is not a service account inherited from the 1990s. Each agent deserves a named identity, permissions bounded to its task, a log of its actions. This can be learned and organised: map the agents already in service, decide who has the right to create them, set what they can commit to on their own and what requires human approval.

GDPR officers know the principle: it is controller accountability. It changes scale when the actor becomes a program that decides on its own, thousands of times a day. Amex and Shopify have just equipped agents to spend. Your teams will use them. What remains is to decide under whose name they will sign.

The Masteria editorial team.

Mathias Nizan, founder of Masteria
Mathias Nizan
Editor of AI Watch
Method

141 articles read, 13 kept

Selected on the morning of 14 October from 38 news feeds. Our method (in French)

Today's sourcesLe MondeLa TribuneNext.ink01netCNILThe VergeNumeramaGoogle DeepMindTechCrunch404 Media

Next edition: Thursday 15 October.

What the 14 October edition means for your teams

Nvidia takes a stake in Doctolib and sells its definition of sovereignty. Does it touch your teams? Let's talk.

Contact us

On compte les visites, pas les visiteurs.

En acceptant, vous aidez un cabinet indépendant à savoir quelles pages vous sont utiles et à améliorer les autres. Aucune publicité, aucune revente, aucun suivi d'un site à l'autre, et vous pouvez changer d'avis à tout moment.