On 17 September 2026, the European Commission presented the EU KIDS Act, a package aimed at a safer digital environment for children and teens. Among its most debated chapters is the treatment of AI companions and conversational chatbots accessible to minors: they must be off by default, must not simulate relationships that create emotional dependency, and should avoid retaining children’s conversations across sessions by default.
Executive Vice-President Henna Virkkunen framed the initiative as safety-by-design. It is not merely fine-print warnings after harm, but a push to stop commercial products from building addictive habits or artificial affective bonds with users who lack full decision-making autonomy.
The proposed text distinguishes general chatbots from companions engineered for emotional intimacy. Both face strict rules when the audience includes minors: pre-launch testing, post-market monitoring, and a real ability to disable the feature. Embedded chatbots could not auto-activate or be promoted toward children.
For under-13s, parental-control emphasis is even sharper. The proposal does not make Europe an AI-free zone for kids, but it imposes deliberate friction: an adult must opt in; a child should not discover an always-on companion.
The virtual-companion app industry—grown on voice, avatars, and long-term memory—reads the KIDS Act as mandatory product redesign. Persistent memory, affection gamification, and daily streaks enter the regulatory danger zone if aimed at minors.
The KIDS Act is not yet law. It must be negotiated in Parliament and Council, a process that may soften or harden age thresholds, definitions of “companion,” and testing duties. What already changed on 17 September is the political baseline: Brussels wrote that AI-mediated emotional dependency is a recognizable harm to children and teens.
Digital-law specialists note interaction with the AI Act and the DSA. The KIDS Act does not replace GPAI duties or very-large-platform rules; it adds a minors-specific layer on conversational products. Firms must map which regime applies to each surface.
Critics fear overreach: does a friendly educational tutor count as an emotional companion. Defenders say the proposal targets features that cultivate attachment and retention, not every school-help interface. The final text will need sharper borders.
For parents and schools, the practical message is dual. More legal tools to demand off-by-default products—and digital-literacy duties that do not vanish, because a default switch does not teach affective manipulation.
Virkkunen and the College place the measure in a summer–autumn 2026 sequence where the AI Office already sent first information requests and European public opinion reacts to cases of minors intensely bonded to chatbots. The KIDS Act turns that anxiety into articles.
U.S. and Asian firms operating in the EU will need localized experiences: a 24/7 companion that works elsewhere may require a radically different “EU minors” mode. That product fragmentation is already a compliance cost, not a hypothesis.
Neural Pulse sticks to the verified core: 17 September proposal, off-by-default for companions/chatbots aimed at minors, ban on addictive/emotional-dependency features, and Virkkunen’s safety-by-design emphasis. Everything else—exact legislative calendar, amendments—remains open.
The implication for product managers is immediate: audit child onboarding funnels, drop chat retention by default, and document non-addiction testing before trilogue sets penalties.
If the KIDS Act survives with teeth, Europe will have defined an exportable standard: conversational AI for minors is not presumed beneficial; it is presumed risky until safe design is shown.
