On September 17, 2026 the European Commission adopted the EU Kids Act proposal, a package meant to stop the online world from remaining a seatbelt-free zone for children and teens. Backed by President Ursula von der Leyen and Executive Vice-President Henna Virkkunen, the initiative places AI chatbots and companions in the same obligation perimeter as social networks, video platforms, and online games.
The demographic core is an access ramp: a ban on social media reaching under-13s and an EU-wide minimum age of 15 to open autonomous accounts. In between, design must be safe by default: no infinite scroll without stopping points, no reward tricks, no push notifications during sleep hours, and no unsolicited contact from strangers. Minors’ profiles would be private by default.
For the AI industry, the companions section most clearly redraws the board. Systems accessible to minors may not use designs that simulate human relationships in ways likely to create emotional dependency. By default, they may not carry a child’s earlier conversations into later chats. They must test for child risks before launch and monitor for harm afterward. Under-13s could use them only through parental-control tools.
If a chatbot is embedded in a platform or game, it may not switch on automatically, may not be pushed at children, and must be easy to turn off. That rule targets a growing commercial pattern: companions that appear as a default “friend” inside educational or entertainment apps, blurring consent.
Virkkunen summarized the philosophy by reversing the burden of proof: platforms — especially very large ones — must show their services do not harm, rather than waiting for regulators to prove injury case by case. For VLOPs, that means prior compliance plans, independent audits at their expense, and an accelerated enforcement regime aiming to close investigations within about 90 days.
The institutional architecture leans on existing Digital Services Act and AI Act structures. The Commission would directly supervise the most widely used platforms and chatbots in the Union; digital-service coordinators and national market-surveillance authorities would cover the rest. It is an attempt to harden substance without inventing a brand-new agency.
Potential penalties align with Europe’s major digital rulebooks: fines that can reach significant shares of worldwide turnover — public debate has cited the familiar 6%-of-revenue reference in sibling regimes — plus orders to pull features. The final text still needs Parliament and Council negotiation, so exact figures may move.
Companion companies argue the rule could chill legitimate educational uses if read maximally. Child-rights advocates reply that educational rhetoric has too often normalized parasocial bonds with systems that owe no parental duty of care. The Kids Act tries to cut that ambiguity with product rules, not only voluntary ethics codes.
For conversational-AI startups, compliance will cease to be a principles PDF: they will need pre-launch tests, harm telemetry, safe defaults, and parental flows. Those selling into Europe without those controls risk faster market blocks than under classic DSA timelines, thanks to express procedures.
It also hits operating systems and app stores, which sit inside the Act’s perimeter. Controlling distribution matters as much as controlling the model: if a companion app cannot list without proving compliance, pressure rises across the whole chain, not only at the lab that trains the model.
The political calendar is deliberately ambitious. Von der Leyen framed online child protection as a mandate priority; the expert panel that prepared the European approach supplies technical cover for the narrative. Interinstitutional negotiation now begins, where platform lobbies and parents’ coalitions will test strength.
Outside the EU, the Kids Act will be read as an exportable template. Regulators in Latin America and Asia who already watch the DSA and AI Act will have a third pillar — minors and companions — ready to copy. Multinationals will prefer one internal global standard aligned with Brussels over fragmenting product by country.
One unresolved tension remains: how to verify age without building identification infrastructure that erodes privacy. The Act speaks of parental controls and age thresholds, yet age-assurance technology remains contested. Regulatory success will depend on not sacrificing one right (privacy) to save another (child integrity).
Neural Pulse reads the Kids Act as the moment Europe stops treating chatbots as “generic software” and recognizes them as high-impact emotional social interfaces. For editors and product managers, the question is no longer whether companions will be regulated, but how much of today’s design survives an emotional-dependency audit.
