The facts
“If there is a risk for children with certain uses of AI, then it must be banned.” In an interview with Les Échos posted on Monday 9 November 2026 in the evening, picked up by AFP, Emmanuel Macron said AI should be banned where it threatens children’s “mental health” and “brain”. “It is not because it is AI that it has to be the Wild West,” he added.
On Tuesday 10 November, at the Paris Peace Forum, a charter of engagement was published by the Élysée. It received the backing of Anthropic, Black Forest Labs, Microsoft and OpenAI. The text is not a law. It speaks of safety by design, the child’s best interests, privacy-respecting age verification, action against child-sexual-abuse material and non-consensual intimate images, and evaluations before putting a system in minors’ hands. An expert committee was due to hand the president recommendations the same day.
The day before, on 9 November, Emmanuel Macron had gathered, in Paris and by video, some fifty heads of state, lab leaders, researchers and experts on the risks of advanced models for collective security. According to his post on X, participants committed to three priorities: better assess those risks, independently and in public; better share alerts after a serious incident; hold those who build models to account. AFP named among the participants the leaders of OpenAI, Anthropic and Mistral AI, and the president of Microsoft. Mistral does not appear among the named backers of the 10 November children’s charter. The applicable legal frame remains the AI Act, whose next useful deadline falls on 2 December 2026.
Editorial view
A charter backed under the cameras does not open a parental setting. It names a problem (age, attachment, content) and leaves it, for now, to the labs’ goodwill. For a parent, the gesture that counts tomorrow morning is not the Forum photograph. It is knowing which assistant runs on a child’s phone, under which account, and who can switch it off.
What this changes in practice. Nothing to configure in the name of this charter. For a small business or freelancer who puts online a service used by minors (tutoring, a game, a chatbot), the text does list the questions a public buyer, a parent or a regulator will ask: age checks, default design, illicit content, evaluation before deployment. For a parent, the right of access and objection remains the concrete lever, not the charter.
Watch-outs. This page is not educational advice and not medical advice. It does not tell a child to use an assistant. The GDPR applies to minors’ data with particular care. The AI Act, on 2 December 2026, in particular bans certain systems used to produce non-consensual intimate content and child-sexual-abuse material: that is a ban, not a charter. On lock-in: a voluntary pledge is withdrawn faster than an article of a regulation.
Trial, or wait. Worth trying: nothing billed as “AI for children” on the strength of this news. Worth doing: check the accounts, the usage times and the accesses already open. Worth waiting for: the public detail of the committee’s recommendations, and any legislative follow-up. The charter itself is already online.
Where it actually helps
Where an adult exercises a right, not where an assistant is put in a child’s hands: prepare a right-of-access or objection request from the CNIL templates, draft a first processing register for a small business. Those pages help frame a service, including if it is one day used by minors. They replace neither a lawyer nor the CNIL.