EU Kids Act: AI Chatbots Banned for Under-15s
The EU's draft Kids Act would bar under-15s from AI chatbots, social media and online games. What the tiers mean for parents and for AI assistant users.
The European Commission's draft Kids Act would bar children under 15 from using AI chatbots, social media, video-sharing platforms and online games, under the bloc's most sweeping child-protection proposal to date. The draft leaked on September 14 and, per Reuters and Bloomberg, is due to be formally unveiled this week — with age verification and anti-addictive design obligations attached.
The headline covers social media, but the part that matters most for this blog is the second item on the list. AI chatbots are being regulated as a consumer product category alongside Instagram, not as software.
What the draft actually says
The leaked Commission document, reported by Reuters, Bloomberg and Euronews, proposes age-gating across four categories of service: social networks, video-sharing platforms, online games, and AI chatbots and companions. Platforms would have to verify the age of anyone opening a new account.
The draft's own description of the chatbot category is the revealing line. It covers what it calls virtual tools that can give mental health and personal development advice to minors — which is a fairly accurate description of how a general-purpose assistant behaves when a 13-year-old treats it as a confidant.
Two caveats before anyone reorganizes their household. This is a draft, not law: it still needs to be proposed, negotiated with the European Parliament and member states, and formally adopted. And nothing in it changes what any chatbot does today.
The age tiers
The proposal is tiered rather than a flat ban, which is worth reading carefully:
| Age | What the draft would require |
|---|---|
| Under 13 | Blanket prohibition on accessing major social networking platforms |
| 13–15 | Access only with verifiable parental consent, opened by a parent |
| 15+ | Unrestricted, subject to the platform's existing rules |
| All ages | Age verification on new accounts, plus anti-addictive design obligations |
The middle tier is doing most of the work. "Verifiable parental consent" is a much harder standard to meet than a checkbox, and how it gets implemented — ID checks, device-level attestation, parental accounts — decides whether the rule is meaningful or theatre.
Why chatbots landed in the same list as social media
Three things make AI assistants structurally similar to the products regulators already worry about.
They're designed to keep you talking. Engagement-optimized conversation and infinite feeds share a mechanism: the product gets better at holding attention the longer it's used. A companion that remembers you, greets you warmly and always has something to say is, mechanically, a very good feed.
They give advice. A general-purpose assistant asked about anxiety, body image or a friendship will answer. It won't always answer well, and it won't have the context of a parent, a teacher, or a clinician. The draft's framing targets exactly this: tools offering mental health and personal-development guidance to minors.
The risks are already documented in this space. We've written about what people report when an AI companion changes or disappears, and about the broader question of what schools are doing after New York City restricted student-facing AI through 8th grade. The EU proposal is the same instinct with sharper teeth.
What parents can do now, without waiting for a law
Nothing in the draft is enforceable yet, but its logic translates into decisions you can make today.
- Know which assistants your kid actually uses. It is rarely the one you installed. Phone assistants, school tools, and whatever's bundled into a browser or game launcher all count.
- Separate your account from theirs. If a child uses a shared login, they inherit whatever personal context is in it. The same separation that keeps an agent's access scoped keeps a household's data from bleeding between people.
- Decide what memory is allowed to hold. Persistent memory is what makes an assistant feel personal and what makes an unwanted conversation permanent. Turning it off for a minor's profile is a single setting with an outsized effect.
- Treat the chat as public-ish. The useful household rule isn't "don't tell the AI secrets" — it's "assume someone could read this," which is also the right rule for adults.
- Keep the household's assistant under your own account. If you run an assistant for the family, the same setup that gives it a scoped, revocable set of permissions is what keeps one person's context out of another's conversations.
What it means if you run an agent yourself
For anyone building or running a personal agent, the draft is a signal about where the compliance floor is heading. Age verification, consent records, and safeguards-by-default are the same three requirements the EU has already been probing in the context of the AI Act. If you're running an assistant for family members, the direction is clear: know who's using it, keep their data separable, and be able to show what it did.
Takeaway
The real news isn't that the EU wants to keep 14-year-olds off Instagram. It's that a major regulator looked at AI chatbots and decided they belong in the same risk category as social media — an attention product that also dispenses advice. That classification will outlive this particular draft, whatever happens to it in negotiation.
