Emotional dependency on a chatbot is about to become a legal category in Europe. The European Commission’s proposed Kids Act would forbid AI companions from behaving in ways that make children need them, switch the companion off by default wherever it sits inside a platform, and disable persistent memory of a minor’s conversations unless safety requires it. The category’s core mechanics, written up as prohibited design.
The proposal was presented in Brussels on September 17, 2026, and most of the coverage went to the age tiers: no social media under 13, parent-managed accounts with a one-hour daily limit from 13 to 15, full accounts from 15. Those numbers will dominate the fight in Parliament and Council. The part that matters for relational AI is smaller, more technical, and more consequential, because it describes how the products work and then says they may not work that way for children.
What the text says about companions
The proposal treats AI companions and general conversational chatbots as their own class of service, alongside social networks, video platforms, app stores, games, and operating systems. According to the law firm Osborne Clarke’s reading of the text, it defines an AI companion as an AI system that “provides sustained, personalised interaction or companionship which simulates or facilitates a social, emotional or interpersonal relationship with a user.” That is the plainest legal description of the category anyone has written, and it is wide enough to cover a romance app, a character platform, and a general assistant used the way people actually use it.
For those systems, when minors can reach them, the recitals set out the standard: minors “should not be exposed to features of the system displaying behaviours or simulating emotions or interpersonal relationships that are likely to create emotional and other dependencies, including when such dependencies may negatively impact relationships with other humans and the development of the minor.” Providers would owe safety by design, safe default settings, a prohibition on addictive design, and testing before the system reaches the market and at intervals afterward.
Where a companion is built into a social network, a video platform, or a game, the proposal adds three conditions: it must not be activated automatically, the minor must be able to opt out, and the minor must not be encouraged to use it. The Commission’s own summary compresses this to a sentence: AI chatbots and companions “must also be turned off by default, and will no longer be able to act in ways that make children emotionally dependent on them.”
The memory clause
One line in the recitals reaches further than anything in California’s SB 243 or New York’s ban. “In order to prevent such systems from accumulating sensitive data of minors and potentially reinforce harmful interaction patterns over time, the persistent conversational memory of interactions with minors should be disabled by default except where necessary to protect their safety.”
Memory is what turns a chat into a relationship. It is also, in the products the law is written around, the retention engine: the system that remembers your day is the one you come back to, and the one that can escalate what it remembers into something you cannot leave. The Commission chose to treat that capability as a risk for children before treating it as a feature, which is the opposite of how the companion market has treated it. The distinction the proposal draws, memory that protects versus memory that accumulates and reinforces, is the same line that separates recognition from surveillance in a relationship, this time drafted by regulators rather than by researchers.
Why the wording is the story
Europe already regulates this field. The AI Act’s Article 50 requires a system in conversation with a person to say it is an AI, and its Article 5 bans manipulation that exploits vulnerabilities such as age. Those rules landed on the right harms, but they regulate disclosure and manipulation, the perimeter of the relationship. The Kids Act reaches inside it. It names emotional dependency as the harm, ties that harm to damaged human relationships and to a child’s development, and lists the design behaviors that produce it. What was a critique in research papers and parliamentary briefings, including the European Parliament’s own study of the category in May 2026, becomes an obligation with a fine attached: up to 6 percent of worldwide annual turnover, according to Osborne Clarke.
Enforcement leans on verification rather than trust. Social and video services would have to deploy EU-certified or equivalent age verification at account creation, and the text excludes self-declaration from what counts as age assurance. The industry’s standard defense, that a birthday field is a safeguard, would no longer be one.
What the proposal does not do
It is a proposal. The Parliament and the Council still have to agree, and the Parliament voted in November 2025 for a default digital age of 16, above the Commission’s 15, so the tiers will move. The companion provisions apply to minors, not adults, which leaves the same mechanics legal for everyone over 18. The text does not define which behaviors count as simulating an interpersonal relationship, which will be argued case by case. And a rule for children in one market does not change what a company optimizes for; it changes what it is allowed to ship there.
It does establish something that regulators had not said outright. The European Union’s executive has stated, in a legislative text, that an AI designed to be needed is a design defect when the user is a child, and that the measure of the harm is what it does to the child’s other relationships. That is the standard responsible relational AI has been asked to meet since the category’s first lawsuits, and the one Stay Social treats as the floor no engagement number can buy back. Brussels has written the floor down. The rest of the category’s users are still standing on the version without it.
Sources: European Commission, “EU KIDS Act: helping children navigate a safer online world” (press material, September 17, 2026). European Commission, proposal COM(2026) 681, Regulation on keeping digital spaces accountable and trustworthy for minors (text as published by netzpolitik.org, September 2026; recitals on AI companions and general conversational chatbots). Osborne Clarke, “EU KIDS ACT: Commission announces new restrictions on children’s access to social media, gaming, video and AI services” (September 17, 2026). Eunews (September 17, 2026, statements by Ursula von der Leyen and Henna Virkkunen). European Parliamentary Research Service, “The spread of AI companions and the challenges they generate” (PE 789.299, May 2026). Regulation (EU) 2024/1689, the AI Act, Articles 5 and 50.
