France's teen social media ban never reached September

On September 1, platforms were supposed to start turning away every new account opened by a fourteen-year-old. They will turn away nobody, because France's Constitutional Council struck the ban down on Friday, August 14, twenty-four days after a vote Parliament had treated as final^[1]. A month ago I wrote about my discomfort with the text, sold on the principle and skeptical of nearly everything else. The Council has now ruled on precisely the point where I got stuck, which ought to feel like vindication and only half does.

The law fell over what it left out

Two referrals, each signed by sixty deputies, followed the vote two days apart, the France Insoumise bloc on July 23 and the Socialists the next day, and the Council ruled a good week before the one-month deadline it was working against. Only Article 1 of the Miller law came before it, the article that carried the entire ban. That article is now unconstitutional.

The decision opens by siding with Parliament on intent, since the Council accepts that protecting a child's best interests can justify limiting what minors reach online. Its quarrel is with the lock rather than the door. A blanket ban applies without regard to any individual minor's situation or to the specific risks of any given platform, so it also catches services "whose risks are not established," and the text set up no way for a parent to authorize access when that serves the child. Because the freedom of communication protected by Article 11 of the 1789 Declaration includes the freedom to reach these services and speak on them, the Council found the restriction neither suitable, necessary, nor proportionate.

Then comes the paragraph I had been hoping for without expecting it. Banning minors means checking everybody's age, including the adults who never asked for any of this, and the Council notes that "the legislature did not provide the legal safeguards" needed to protect the privacy of everyone forced to identify themselves. That was half of my worry back in July, the half that had nothing to do with teenagers and everything to do with our ID photos, and it now sits in a ruling that binds every branch of the French government.

A green light that was never binding

Laure Miller, the deputy from the Marne who wrote the bill, responded with a line worth sitting with, "I have a hard time understanding it," explaining that she had not expected the ruling because the Council of State, the government's in-house legal adviser, had told them a blanket ban would survive constitutional review since this was a public health matter^[2]. That argument would have carried if public health had been in dispute, and it never was for a second. The science is exactly where it stood in July, the French food safety agency's review of 11-to-17-year-olds and the microsimulation published in PLoS Medicine have not moved an inch, and the Council left both standing when it endorsed the goal. Parliament argued the legitimacy of the destination in front of a judge who was only examining the route.

Servane Mouton, a neurologist who co-chaired the national commission on children and screens, sees the ruling as the burial of an urgent protection against platforms she calls toxic^[3]. I understand the anger without sharing the reading, since a law struck down for missing safeguards is a law you can rewrite, provided you admit you rushed it. The Council did not say to give up. It said to write down the how, which happens to be the title of my last piece and took no prophetic gift, given the size of the hole.

What survives and what starts again

Not everything fell. The ban on phones in high schools, tucked into the same law, appeared in neither petition and takes effect when classes resume, so the least argued provision in the text will be the only one students actually meet on September 1.

Everything else moved fast. Within hours, Emmanuel Macron had instructed prime minister Sébastien Lecornu to prepare a "legally sound" version as quickly as possible, aiming to land it by spring 2027 and to square it with the European framework^[4]. Miller intends to rework her own draft. The timeline is worth a laugh, since a bill carried for eighteen months and struck down in three weeks now restarts for another eighteen, this time in an election year, and an election year has never improved anyone's legislative drafting.

The ruling did hand the government something it lacked, which is a specification. A second version holds up if it varies by age and by service instead of closing the social web in one gesture, if it lets a parent allow what that parent judges allowable, and above all if it writes into law the verification mechanism the first version forgot. That mechanism already exists. The standard published by Arcom, France's broadcast and digital regulator, and cleared by the data protection authority has governed adult sites since January 2025 through double anonymity, where the site learns you are over eighteen without learning who you are, while the verifier learns who you are without learning where you went^[5]. It took a constitutional defeat to make that document legible to Parliament.

Meanwhile, California shipped

On August 18, four days after the ruling, OpenAI launched ChatGPT for Teens, a version of its assistant built for thirteen-to-seventeen-year-olds. It opens in waves and should reach the rest of the world within about two weeks^[6]. A teenager lands there automatically, either because they stated their age, or because their account details give them away, or because the system guesses they are under eighteen from the kind of questions they ask. When it cannot tell, it assumes minor and leaves the adult to prove otherwise.

The line that matters for my purposes sits in the help documentation. OpenAI receives no ID document and no photo of anyone's face, only a date of birth or an age estimate, while the rest stays with the vendor doing the checking^[7]. That is the Arcom standard, adopted voluntarily by an American company nobody had asked. On content, the assistant pulls back on self-harm, suicide, and anything romantic or sexual, and refuses to imply it feels anything, a precaution aimed less at modesty than at attachment. On the parental side, both parties have to opt in, parents can set quiet hours and get alerted when their kid appears to be in acute distress, and at no point do they get to read the conversations.

Technically that is close to the Council's list of complaints, handled in order and shipped in a week. It grades access by age instead of closing the door, it verifies without collecting papers, and its parental controls stop short of surveillance. I would be dishonest if I pretended not to enjoy this, having asked for all of it a month ago.

I would also be dishonest if I stopped there. Estimating age from the questions someone asks means reading what they write on a running basis, which trades the ID card for behavioral profiling, and that trade costs something. A safeguard no parliament voted on can vanish the day its author reconsiders, with no debate and no appeal, whereas a law that gets struck down can be written again. A private company is drawing the boundary of childhood on its own, using criteria it publishes when it feels like it and revises whenever it wants.

I walked away with both worries intact

Nothing has shifted in my position since July 24, which is news in itself. I still want teenagers protected and still refuse to hand over my papers to get there, except that those two demands now have a referee, and it ruled against the text on the second without ruling against anyone on the first.

Spring 2027 will show whether somebody read the decision all the way through or only the press release. Until then, the only protection actually deployed for French teenagers will have been written in California by a company nobody elected, which can withdraw it on a Tuesday morning without telling anyone. That is not the kind of win I was after.