On 14 August the French Constitutional Council struck down the law banning social media for under fifteens. Parliament had passed it on 21 July and it was due to take effect on 1 September for new accounts. The reasoning is worth reading twice: a blanket ban is disproportionate to freedom of expression, and there are not enough safeguards on age verification or on the privacy of people made to identify themselves.

The second point matters more than the first. To enforce an age ban you have to know everybody’s age, not just the minors’. It means building a system in which every adult proves who they are in order to open an app.

The French government has already announced a sturdier text. And this is not a French accident: Denmark has had a political agreement since November 2025, Spain has a bill before parliament, Greece has announced a ban from 1 January 2027, the United Kingdom published its response in June to a consultation that drew 116,000 replies. The European Council discussed it on 19 March, the G7 on 17 June. In Italy, bill 1136 has been stuck in Senate committee since 21 October 2025.

Of all these countries, exactly one has a law in force.

Australia actually tried it, and the numbers exist

Australia’s ban for under sixteens has been in force since 10 December 2025. It is the first in the world. On 31 July 2026 the regulator that oversees it, eSafety, published its three month assessment, based on a survey carried out in March.

Before the ban, 85.9 per cent of under sixteens said they used at least one restricted platform. Three months later: 81.5 per cent. The share using them daily or more often went from about 60 to 58 per cent. Roughly half of those who kept their account say the platform never checked their age.

Four and a half points on use, two on daily use. That is the result of the strictest law ever written on this subject, in a country with one language, one regulator and no land borders.

An independent study published in the British Medical Journal in June 2026, following 408 adolescents aged twelve to seventeen in New South Wales before the law and three months after, reaches the same conclusion in more careful language: insufficient evidence of substantial early effects.

The limits are worth stating too, otherwise we are in the same business as the people selling the ban. Three months is short. The figures are self reported. The BMJ study covers one state and measures use, not wellbeing. And in June the Australian government doubled the penalties on platforms, which means it regards the first half as over rather than the match.

The serious objection, met head on

The strongest reply to all this does not come from governments. It comes from the experts who support the ban. The special panel on child safety online set up by the European Commission, in its report of 13 July, acknowledges that most minors circumvent bans. Then it adds that social norms change over long horizons, and draws a precise parallel: attitudes to the ban on corporal punishment in schools took more than a generation to shift, in Germany and in Sweden.

It is a real objection. Judging a law meant to change a custom after three months is like judging that ban in its third month.

But there is an answer. Nobody promised a generational shift. The Australian law promised to remove access, immediately, and access is the only thing it could measure at three months. On that it failed, and it failed for a technical reason that no amount of time fixes: the platforms did not check ages. That is not a matter of custom. It is a matter of enforcement.

Where the two sides start to look alike

The same European report contains a sentence worth more than the rest: the burden of proof must sit with providers, not with regulators, parents or children. And it makes clear that the under thirteen restriction is designed as a precautionary measure on platforms, one that lifts when they demonstrate they are safe by design.

Put that way, it is not offloading responsibility onto minors at all. It is a reversal of the burden of proof, and it is a liberal argument.

Which raises a different question. If the test is putting the burden on whoever builds the service, that instrument already exists and it is not a ban. Article 28 of the Digital Services Act requires platforms accessible to minors to put safety measures in place and forbids profiling based advertising aimed at them. European data protection law carries fines of up to 4 per cent of global turnover. Both have been in force since 2022 and are barely used.

An age ban is easier to pass than a case against a platform worth a trillion dollars. That is where the shortcut is.

Estonia, which said no

On 10 October 2025 twenty five EU states plus Norway and Iceland signed the Jutland Declaration, a joint commitment to restrict minors’ access to social media. Two did not sign: Belgium and Estonia.

The Estonian position, argued by both the education ministry and the ministry of justice and digital affairs, makes three points: bans get circumvented, they push young people towards smaller and less supervised platforms, and they take the pressure off whoever created the problem.

And it offers the alternative Estonia has been running for fifteen years. This comes from the Estonian education ministry’s own national media literacy report of December 2025, not from a newspaper account. Since 2011 there has been a compulsory 35 hour course, “Media and Manipulation”, in upper secondary school, extended to Russian language schools in September 2023. Since 2023 developing media literacy is a requirement of the national curricula for primary, secondary and vocational education, and digital competence is one of the general competences every teacher must build into their subject. From September 2025 it reaches early childhood education. Also from September 2025 a national programme on artificial intelligence in schools began, with twenty thousand students and three thousand teachers in its first year.

The evidence almost nobody brings

So far this is a description of a country doing a thing. But a country doing a thing is not evidence that the thing works, and honesty requires saying so: the Estonian ministry writes plainly that impact evaluation is hampered by the absence of transparent metrics. Estonia itself cannot measure the results of its own model.

The evidence does exist, and it comes from experimental research. A meta analysis published in Communication Research in 2024 pooled 49 experimental studies covering 81,155 people: media literacy interventions produce an effect of 0.60 standard deviations on resilience to misinformation, 0.76 on the ability to tell true from false, 1.04 on reducing the sharing of false content. A second meta analysis, on adolescents, reports an effect of 0.48. Repeated interventions work better than one off sessions, and they work better when guided and delivered inside school.

In plain terms: teaching works, and we know it with the same kind of evidence that tells us a treatment works. The ban, by contrast, has exactly one completed experiment in the world, and it produced four and a half points.

Where the liberal point sits

This is not an argument that social media is good for teenagers. It does measurable harm, and people who support the ban are not naive: 92 per cent of Europeans consider protecting minors online a priority, and they have their reasons.

The point is that faced with a risk there are two roads. One removes freedom from people not yet able to use it and returns it on a fixed date, handing an eighteen year old a world they have never seen, all at once, with nobody watching any more. The other gives them the tools while the freedom grows, and holds to account whoever builds the environment in which that freedom is used.

The first is easier to draft and shows results immediately. The second is slow, it costs money, and the results arrive when whoever decided it is no longer in office. It is the only one that produces adults.

And there is something a ban does quietly, which is the worst of it: it tells a young person that their freedom is a problem to be managed rather than a thing to be learned. People who grow up hearing that, later, hear it about other things.

What remains open

Estonia is not a digital paradise, and it is better to say so before somebody else does. In the World Health Organization’s survey of adolescent behaviour, Estonian boys are among the most at risk in Europe for problematic gaming: nineteen per cent at eleven, seventeen at thirteen and at fifteen. Only two countries do worse. A curriculum is not a shield.

Estonia is also a country of one point three million people, with a uniform school system and the most digitised public administration in Europe. Anyone claiming we should simply copy it is selling a shortcut, exactly like the people selling the ban. Italy sits twenty third of forty one in the European index of vulnerability to disinformation: a middling starting point, though a rising one.

The right question is not whether to copy Estonia. It is why Europe is legislating on the basis of the one experiment that has already produced numbers, while ignoring them, and why in four years almost nobody has asked about enforcing the rules we already gave ourselves.

Banning is easier than teaching. And it costs less, right up until the bill arrives.