Non, a Ban Is Not a Plan: France Struck Down the Ban, Meta Settled on Features
For some background music when reading this post, we encourage you to (re)discover Michel Polnareff’s “La poupée qui fait non”.
France adopted its under-15 social media ban on 21 July 2026 but hit a major roadblock a few weeks before its prohibition was due to take effect. The Conseil constitutionnel censured the central provision of the ban on 14 August, and the reasoning for its “Non” travels far beyond France. While the court did accept that protecting the youngest minors can justify limiting their access to online social network services, it then struck the provision down because the legislator had set out to address the risks posed by certain features of those services and instead wrote a rule about the age of the user.
Four grounds carried the censure, and each one is interesting beyond the specifics of this case. The published decision finds the prohibition general and absolute, applying to every minor under fifteen without distinction according to the type of platform or the risks each one presents, a disproportionate interference with the freedom of expression and communication protected by Article 11 of the Declaration of the Rights of Man and of the Citizen. It cut against the best interests of the child on the same logic, since parents were left unable to judge whether a given service suited their own child, and it failed on privacy, since the law required every user, minor and adult alike, to have their identity and age verified without ever fixing the conditions under which that verification would happen. Finally, the court found that the scope of the prohibition and the penalties for breaching it were too imprecise to meet the constitutional requirement that citizens understand the law that binds them.
On 16 September, the President of the European Commission, Ursula von der Leyen, will address Parliament in Strasbourg for her yearly State of the Union. She promised repeatedly an instrument related to minors and social media for after the summer, and the expert panel that reported on 13 July recommended a harmonised restriction on access for children under thirteen. Will the lessons from France impact what comes out of Brussels?
The French decision makes for an interesting drafting brief for Brussels
What are the takeaways from France’s Constitutional Court that could be helpful in moving this debate from the question “Is a Ban a Plan” that we examined in the first half of this year, to “So, what is a plan?”.
First, distinguish services by what they do. The French law exempted online encyclopaedias, educational and scientific repositories and open-source educational projects, and the court observed that those carve-outs left collaborative content-sharing services, online communication applications and games with pronounced social features inside a prohibition that had never been justified by reference to them. The prohibition could therefore catch services whose risks to the health and safety of minors, arising notably from their content or their manner of functioning, were not established.
Second, leave room for the individual child and for the adult responsible for that child. The panel advising the Commission seemed to take the opposite position, insisting that primary responsibility must sit with providers, not with parents or minors. Both propositions can hold, but only in an instrument that regulates providers and still leaves parents something to decide, since a statutory access age gives a provider a compliance task and gives a parent a fait accompli.
Third, specify how age is to be established before requiring it. Availability of an age-verification method still leaves open who verifies, what is retained, and what a wrongly classified user is entitled to, and a statute must fix all three or fail on privacy the way the French one did.
Fourth, and this is where the ruling bites hardest on Brussels, define the thing being prohibited. EU law contains no binding definition of social media. The closest text is the online social networking service in Article 2(7) of the Digital Markets Act, a definition built for gatekeeper designation, and France borrowed from it for want of anything better. The Digital Services Act (DSA) defines online platforms and online search engines, the Audiovisual Media Services Directive (AVMSD) defines video-sharing platform services, and the AI Act defines AI systems. A prohibition assembled from those pieces, or from a new category invented to sit across them, invites precisely the intelligibility objection the French law failed to survive.
Twenty-three countries in the EU are envisaging building the same defect
As of May 2026, twenty-three of twenty-seven member states were at least contemplating national legislation, and the results do not converge. The proposed ages run at fourteen in Austria, fifteen in France, Denmark, Greece and Italy, and sixteen in Spain and Poland, with Norway proposing the same from outside the Union. The same fifteen-year-old is old enough in Paris, too young in Madrid, and a matter for non-binding guidance in The Hague, without changing anything but her location. The scopes diverge further still: Slovakia reaches AI services and loot boxes, Portugal reaches app stores and betting services while carving out the main messaging applications, and the Netherlands has issued guidance with no statutory force at all, three different answers to the same question of what falls inside the perimeter. The Commission’s own panel widened that perimeter again by coining the phrase social media plus to pull gaming, messaging and AI companions into the same net. Enforcement diverges too, split between national identity wallets, platform-side obligations and ministerial blacklists, and the politics divide along the same lines: Belgium declined to sign the Jutland Declaration, Estonia and the Czech Republic oppose bans outright, and Germany has spent a year commissioning assessments without producing a bill.
The European Parliament identified the consequence before France supplied the constitutional argument for it. The resolution on the protection of minors online, adopted on 26 November 2025 by 483 votes to 92 with 86 abstentions, records that divergence in age assurance fragments the internal market, and calls for a harmonised digital minimum age of sixteen with parental consent from thirteen. That resolution, however, binds nobody, and it points to the Digital Fairness Act, a consumer-protection instrument scheduled for the fourth quarter of 2026 whose subject matter is dark patterns, addictive design and unfair personalisation, with no statutory access age initially planned in it. The single-market case for acting at EU level is sound. A national court may have weakened the case for the particular instrument everyone assumes is coming.
The Australian data removes the fallback argument that it works
The reply to a proportionality objection is usually that the measure works. Australia has now run the experiment long enough to test that.
The eSafety Commissioner’s three-month evaluation, released in July 2026, found that use of at least one age-restricted platform among ten- to fifteen-year-olds fell only from 85.9 per cent before the ban to 81.5 per cent after it. Account ownership fell further, from around 52 per cent to around 42 per cent, but most children who held an account before commencement kept it, and the commonest reason they gave was that the platform had never asked them to confirm their age. eSafety is careful to say this is early, and that platforms have not yet done the work the law expects of them. That caveat cuts both ways. It is why the numbers are so modest, and it is also why the usual promise, that compliance will build once platforms act, deserves real testing rather than assumption.
A National Bureau of Economic Research working paper by Cass Sunstein and colleagues at Chicago, Pennsylvania and Harvard did conduct some testing. Surveying Australian teenagers, they found roughly 27 per cent of banned fourteen- and fifteen-year-olds complying four months in, with three quarters describing circumvention as easy or very easy and more than a fifth of non-compliers reporting that an adult had helped them sign back up after a deactivation. The key finding is not the percentages but the direction. Teenagers said they would need about two-thirds of their peers to leave the platforms before they would leave themselves, and they regarded the ones who complied as less popular, not more. Being the child who goes quiet while the room keeps talking is a cost rather than a badge, so compliance of this kind erodes rather than builds, and by a June follow-up it had fallen to about 6 per cent, even as platform enforcement tightened rather than loosened. The authors are careful, noting that a ban might still work for a future cohort that never joined social media in the first place, and that the answer is to pair any restriction with limits, incentives, and norms rather than to prohibit and hope. It has not yet been peer-reviewed, but the compliance numbers are echoed by surveys by the Molly Rose Foundation and YouGov.
Use also migrated toward services the ban did not cover, so, as anticipated by experts, the ban did not remove these children from social media. It removed them from its safer, regulated version.
Two earlier information labs posts, Don’t Be Fooled and Panic First, Evidence Later, set out why the Australian model travels badly to Europe and why the evidence base was thin before anyone legislated on it. Sonia Livingstone made the same point about what the evidence can and cannot carry in her contribution to the repository. A measure that restricts a right, barely dents use, and then loses even that ground as compliance collapses is the weakest proportionality case available to a drafter. The French court struck down a ban that had not yet taken effect. An EU ban would arrive with Australia’s numbers already attached to it.
The Commission’s own DSA cases already sit where the court put the harm
Enforcement under the DSA has been moving toward design for two years, and it has been moving successfully. The guidelines under Article 28(4), published in July 2025, were followed in October 2025 by information requests to Snapchat, YouTube, the Apple App Store and Google Play. The preliminary findings against TikTok of 6 February 2026 concern infinite scroll, the absence of screen-time interruption and a recommender weighted toward engagement. The preliminary findings against Meta of 10 July 2026 concern infinite scroll, autoplay, push notifications and the same recommender logic, with the Commission stating that the recommender should be made less engagement-oriented. Proceedings opened against Snapchat in March 2026 address age assurance and design enabling unsolicited adult contact. Article 28(2) already prohibits advertising to minors based on profiling. None of this amounts to a final decision, but together it is the strongest position the Commission currently holds. Each of those cases identifies harm as stemming from a feature. So did the French legislator, before deciding the threshold was an age.
Meta has now agreed to change the features, not the age
Within a fortnight, the largest of those companies agreed to the same reading. On 26 August, twelve days after the French ruling, Meta settled claims brought by forty-seven United States attorneys general over the effect of its platforms on young users, and the remedy it accepted is a set of design changes rather than an age bar. Under the proposed agreement, which still requires court approval, users under eighteen on Facebook and Instagram in the participating states would default to a feed that is not personalised, gain a switch to turn autoplay off, lose push notifications during school hours, meet a hard daily time limit with pauses built in, and find the apps blocked overnight, while like counts are hidden and the harshest appearance filters are removed. An independent auditor would verify that the measures are built and that they work, and most are set to hold for ten years.
These are, almost one for one, the features the Commission named in its preliminary findings against the same company on 10 July, and several sit among the design problems the repository sets out to fix. The attorneys general called them addictive by design, which is the regulator’s language rather than ours. Meta conceded that the harm sits in the features, and that the features can be changed. The settlement still turns on knowing which users are minors, which is why it also brings in stronger age assurance, and that is the third of the French court’s requirements rather than a departure from it: settle how age is established, then attach the duty to the service instead of to the child.
Four principles hold a valid plan together
None of this needs to be invented from scratch. Read the four grounds above against the Commission’s own findings on TikTok and Meta, and the outline of an instrument is already visible, the same outline the revamped Social Media Ban Repository organises its recommendations around under four headings. Prove it safe before it ships comes first: digital products tested against standards before they reach the market, the way physical products already are, with the burden of demonstrating safety sitting on the provider, not on a regulator who has to catch the problem afterwards. Second is Child Rights by Design, which applies that standard without regard to category, so games, chatbots and EdTech answer to the same rules as social media: safe and private by design and by default, the same distinction the French court drew when it faulted a law for exempting the wrong services for the wrong reasons. Third is regulate the feature, not the user, the ground the decision supplies directly: the obligation targets the design practices that generate harm rather than the children who navigate them. The fourth heading, co-designed with young people, treats children’s own perspectives as indispensable intelligence for that design work, not an optional supplement gathered once the drafting is already done.
Those four principles answer France’s four grounds together rather than one at a time. They distinguish services by what they do, they leave a parent an actual switch to operate and a child a capacity to grow into rather than a birthday to wait for, and they ask nobody to establish anybody’s age, since they are drafted from definitions already in force across the DSA, the AVMSD and the AI Act, not from a category the French law never managed to define. Written this way, they also leave the enforcement pipeline intact, codifying the standard the Commission is already applying to TikTok and Meta rather than handing either company a reason to argue that a newer, more specific text now governs the same conduct.
The road to the State of the Union: paved with good intentions or solid principles?
When the President of the Commission addresses Parliament on 16 September, one month after the ruling, the instrument capable of surviving judicial review is the one built around design: the feed that never ends, the video that plays itself, the streak and the reward loop, the notification built for return, the feed that optimises for time, the stranger in the inbox, the profile built to sell, the report that goes nowhere, the broadcast that cannot be taken back, the map that knows where you sleep, the box you pay to open and the companion that never logs off. Each of those twelve is definable in existing EU law. None of them requires knowing anyone’s age. None of them requires invading everyone’s privacy.
We have spent the first half of this year documenting what bans do. From the collected evidence, we identified what a plan could look like. The revamped repository is live at informationlabs.org/social-media-ban-repository, setting out the twelve design problems, the fix each one takes, and the sources proposing every fix. It answers the question the repository opened with in Is a (Social Media) Ban a Plan?.
The doll in the Polnareff song shakes its head because that is all it was built to do, and Polnareff complains that it does not listen or watch him at all. A blanket ban is the same reflex. A plan has to say more than “non”. And a plan affecting the young generation should be built with them, not against them.
Written by Caroline De Cock, LL.M., Head of Research
