Confirmed in a primary source 17 Sept 2026 Policy

Europe proposed a law that bars under-13s from social media and puts AI companions under children's safety rules.

On September 17 the European Commission adopted the EU KIDS Act, a draft regulation that would stop children under 13 from holding social media accounts, limit 13 and 14 year olds to accounts run through a parent's, set 15 as the age at which a child can open an account alone, and apply safety rules to AI companions and chatbots offered to anyone under 18. It bans a list of specific product features for minors: recommendation feeds built on profiling, endless scrolling, reward loops, notifications at night, messages from strangers, and AI companions that simulate a human relationship by default.

The caveat that matters is that this is a proposal, not a law: it now goes to the European Parliament and the Council, either of which can change the age thresholds, narrow the scope or drop the AI provisions before anything binds anyone.

Until now

AI companions were regulated, if at all, through general rules on platforms and through the AI Act's transparency duties, with age limits set by each company's own terms of service.

Now

Treat the AI companion as a named product category in European child safety law, banned by default for under-18s in the Commission's text, with the burden on the provider to prove otherwise.

Yesterday the President of the European Commission stood in Strasbourg and said it was time to slow down on the self-recursive models. Today her Commission adopted the first of the instruments she named. The EU KIDS Act does something the speech only implied: it treats an AI product as a children's safety problem in the same breath as a social network. Under-13s get no social media account. Thirteen and fourteen year olds get an account their parent holds the keys to. Fifteen is the age you can sign up alone. And AI companions, the chatbots built to act like a friend rather than a search box, are inside the scope for everyone under 18, with the ones that simulate a human relationship switched off by default.

The mechanics are where the weight sits. The proposal bans named features rather than stating principles: profiling feeds, infinite scroll, reward loops, night-time notifications, strangers messaging children. It sets fines at 6 percent of global annual revenue, the same ceiling as the Digital Services Act, and makes platforms pay a fee for the cost of supervising them. It gives the Commission 90 days to finish an investigation, which is fast by Brussels standards and is the number to watch, because an enforcement deadline is a promise a regulator makes about its own behaviour. What it does not do is bind anybody yet. This is a draft. The Parliament and the Council will spend the next year arguing about the age thresholds, and the AI provisions are the newest and therefore the easiest to trade away. Treat the AI companion as a named product category in European child safety law, banned by default for under-18s in the Commission's text, with the burden on the provider to prove otherwise.

How it got here

Aug 2023
The Digital Services Act begins applying to very large platforms, setting the 6 percent revenue ceiling this proposal reuses.
Aug 06
The main obligations of the EU AI Act come into force, with transparency duties for chatbots but no age rules.
Sep 16
Ursula von der Leyen tells the European Parliament it is time to pace the frontier, and names the Kids Act as one of the instruments. Carded in Edition 024.
Sep 17
The Commission adopts the EU KIDS Act with AI companions and chatbots in scope, and sends it to the Parliament and the Council.
Who has to react
Every company running an AI companion product in Europe, Character.AI and Meta's persona products first among them, which now face a default-off rule for minors rather than an age gate they write themselves; OpenAI and Anthropic, whose assistants are not companions but whose under-18 users are covered by the safety-by-design duties; Meta, TikTok and YouTube, for whom the 13 and 15 thresholds are a product change in every member state at once; the games industry, in scope for the first time in a child-safety instrument; the European Parliament's IMCO and LIBE committees, which now own the text; national data protection authorities, whose age verification quarrels this centralises; and the American administration, which has treated European platform rules as trade measures.

The evidence

ObservedThursday September 17, 2026. The European Commission adopted the EU KIDS Act, where KIDS stands for EU Keeping Internet Digital Spaces Accountable and Trustworthy, and published it as press release ip_26_1890 with the proposal text and FAQ posted to the Commission's digital strategy library the same day.

The whole observation, as recorded

Age thresholds: no independent social media account under 13; 13 and 14 year olds restricted to limited accounts accessible through a guardian's account and to video services designed for children; 15 as the EU-wide minimum age to open an account alone; safety-by-design obligations for every user under 18. Scope covers social media platforms, video-sharing services, online games, and artificial intelligence companions and chatbots offered to under-18s. Banned for minors: addictive design features, profiling-based recommendation feeds, infinite scrolling, reward mechanisms, notifications during sleeping hours, unsolicited contact from strangers, and AI companions simulating human relationships by default. Penalties reach 6 percent of global annual revenue, and platforms also pay a supervisory fee covering the cost of their own oversight. The Commission gives itself 90 days to conclude a fast-tracked non-compliance investigation. Large providers must file compliance plans with the Commission and put new products past an independent auditor. Age assurance runs on the EU's own open-source age verification solution, due by the end of this year. Ursula von der Leyen: Our KIDS Act is reversing the burden of proof, it is for platforms to show they are safe by design. The proposal goes to the European Parliament and the Council. This is the instrument named in Edition 024's Europe card, adopted one day after the speech that announced it.

Established

The European Commission adopted the EU KIDS Act on September 17, 2026, and published it as ip_26_1890 with the proposal and FAQ on the same day.
The text puts artificial intelligence companions and chatbots offered to under-18s inside its scope, alongside social media, video-sharing services and online games.
It sets age thresholds of 13, 15 and 18, a penalty ceiling of 6 percent of global annual revenue, a supervisory fee, and a 90 day deadline for fast-tracked investigations.
It reverses the burden of proof, requiring providers to show their services are age-appropriate and safe by design.

Not established

That any of this is law. It is a Commission proposal at the start of the ordinary legislative procedure, and both co-legislators can change it.
That AI companions will be banned for minors in Europe. The default-off rule is in a draft that the industry will lobby against for the next year.
That the pacing argument from yesterday's speech is being applied to frontier model developers. This instrument reaches consumer products aimed at children, not training runs.
That enforcement will match the text. The Digital Services Act has the same penalty ceiling and has produced few completed cases against the largest platforms.

Why it weighs this much

Signal strength

CostModerate

CommitmentModerate

ReversibilityMedium

PersistenceYears

Cost is moderate and real: adopting a College text takes an impact assessment, legal service sign-off and the political capital of picking a fight with American platforms in the same week as a trade argument. Commitment is moderate rather than high because a proposal binds the Commission to defend a position, not to deliver an outcome. Reversibility is medium: the Parliament and the Council can rewrite any of it, and the AI provisions are the newest and least defended. Persistence is years because European files move slowly and because the thresholds, once set, become the reference point every platform builds to. We are not scoring this as a law. We are scoring it as the first text in which an AI companion is a regulated children's product.

What to watch

strengthens The European Parliament's lead committee adopts a report keeping AI companions in scope with the default-off rule intact, or the Council's general approach does, by December 31, 2027.
weakens The AI companion and chatbot provisions are removed or reduced to a transparency duty in either co-legislator's position, or the file stalls with no committee vote, by December 31, 2027.

The words

EU KIDS Act: The Commission's draft regulation on children online. KIDS stands for EU Keeping Internet Digital Spaces Accountable and Trustworthy.
Safety by design: A rule that a service must be built so the harm cannot easily happen, rather than policed after it does. Here it means the provider has to show the design is safe, not the regulator show it is not.
AI companion: A chatbot built to hold an ongoing personal relationship with the user rather than answer questions. The proposal names these separately from ordinary chatbots.
Proposal: A draft law the Commission writes. It becomes binding only after the European Parliament and the Council agree a final text, which usually takes one to two years and changes the draft.

Your call

Which explanation do you believe?

A call is final. Other people's calls show after yours.