The EU's draft Kids Act would strip companion chatbots of emotion and memory for under-18s

The draft goes at the two features that make a companion product work. One is continuity: the bot knows what you told it last Tuesday ^[4]. The other is emotional register: it responds as though it has feelings about you ^[1]. Around half of Europeans aged 11 to 25 have discussed intimate or personal matters with an AI chatbot, according to an Ipsos BVA survey across France, Germany, Sweden and Ireland ^[7]. Seven of the fifteen single-year cohorts in that band sit under 18 and inside the proposal's scope ^[8].

The Commission's stated reason for the memory rule is data accumulation. It worries that systems will build up sensitive data about minors over time and reinforce harmful interaction patterns ^[5]. It also says dependency may hurt a child's development and their relationships with other people, and that chatbots give minors dangerous advice on mental health and personal development ^[14]. "Platforms have been playing fast and loose with children's mental health," von der Leyen said at a press conference on Thursday ^[11]. As WIRED describes the memory rule, it applies to chatbots operating in the EU, not only to accounts held by minors ^[4].

Inside teams, memory is the thing that removes repeated setup and makes an assistant useful in week four instead of week one. The draft treats it as a file that keeps growing on a 14-year-old. The one outcome measure in the Commission's case runs against the feature. Across 91 countries, children who say they use AI chatbots for schoolwork get lower test scores than children who do not, according to OECD research ^[9]. The OECD figure compares two groups of children; it does not show what causes what.

The proposal also goes after defaults. Social networking sites, video-sharing platforms and games should not automatically activate or encourage AI chatbots, and should not pin AI companions or general conversational bots at the top of a minor's contacts list ^[10]. Snapchat had done that ^[10].

Both age rules turn on knowing who is 12, who is 17 and who is 19. WIRED's account of the draft does not say how a platform is meant to establish that. The determination falls to whatever age signal a team already collects, and the build follows from it: a session-scoped mode with memory writes disabled, plus a response policy that strips first-person affect, applied to every account the system believes belongs to a minor.

The scoping call belongs to each surface separately, and it has two parts: can someone under 18 reach it, and does it hold state between sessions or talk about its own feelings. An order-status bot that forgets you answers no twice and changes nothing; a companion answers yes twice. The hard case is the general assistant with memory on by default that a 15-year-old can sign into, because the draft only reaches chatbots the Commission counts as posing serious risks to minors ^[6].

Simeon de Brouwer, a policy adviser at the Brussels-based advocacy group European Digital Rights, welcomed the chatbot measures but told WIRED that "it doesn't make sense to restrict them to young people only" ^[12]. The EU says its Digital Fairness Act, due later this year, is expected to propose clamp downs on addictive design more widely ^[13].