EU Kids Act strips the feature that made Character.AI work for teenagers

Character.AI and Replika are not allowed to remember what teenagers tell them. The EU Kids Act — the European Commission’s proposed digital safety regulation, now released as a full legislative text — requires AI companion applications to delete the memory of any conversation with a minor once it ends, unless the data is directly needed to protect the user’s safety. The design that made these apps culturally significant to millions of teenagers is now, under European law, prohibited.

The companion provisions go further than a memory wipe. Platforms must disable emotional companion features by default for all users under 18. Any design element that simulates an interpersonal relationship, fosters emotional attachment, or rewards frequent engagement is explicitly banned. The Commission named the exact mechanisms these apps use to retain teenagers: conversational continuity, the experience of being known, the sense that the AI is invested in the user’s life. All of it must be removed.

The act covers substantially more than AI chatbots. Under-13s are blocked from social media platforms, online games, and AI applications without parental authorization. Autoplay, infinite scroll, and engagement notifications during sleeping hours are prohibited for all users under 18. Geolocation, microphone access, and cameras must be disabled by default. Self-declaration of age no longer satisfies verification requirements; platforms must integrate certified EU age-verification systems.

Character.AI has already announced parental controls and session time-limits, moves made under mounting regulatory pressure before the Act’s formal text appeared. Replika committed to building “companion boundaries”: the AI will tell users it cannot substitute for human relationships. The larger platforms — Meta, Google, Snap, OpenAI — have offered no formal public statement. Compliance analysts estimate significant redesign costs for all of them. Meta, which recently settled a $17.1 billion lawsuit across 47 US states over allegations it designed platforms to addict children, faces compounded exposure under the new rules.

The skepticism case is substantial. The Kids Act is a proposal, not law. It enters months of negotiation between the European Parliament and member states before becoming binding, and its final provisions are likely to change. Genuine age verification at the scale of Instagram or TikTok remains technically hard. Building memory-deletion into AI companions architected around continuity is a deeper engineering challenge than regulators may have anticipated. Companies already note that companion chatbots do not retain memories across separate sessions in all implementations, making the ban’s scope unclear in practice.

The broader context matters for readers outside Europe. Australia prohibited social media for under-16s late last year. Forty-seven US states reached a combined $17.1 billion settlement with Meta over child addiction allegations, and multiple US states have laws restricting minors’ social media access. The EU Kids Act is the most aggressive single proposal so far — its particular target is not access but emotional design itself.

The European Commission published the full legislative text on September 19. Parliamentary negotiations open in October. Once implemented, companies will have six months to identify existing accounts belonging to users under 15 and either convert them to parental-control models or deactivate them. Violations carry fines of up to 6% of annual worldwide revenue.

Tags: AI regulation, European Commission, Character.ai, AI chatbots, children digital safety, EU Kids Act