EU AI Companions and Minors: The Plan to Switch Them Off

On September 17, 2026, the European Commission proposed a law that treats chatbot memory the way many parents already worry about it: as a risk for children. Under the proposed EU KIDS Act, providers of AI companions and general chatbots would have to ensure that “by default, their system does not use information or analysis derived from a minor’s prior interactions in subsequent interactions.” That is the core of the EU’s plan for AI companions and minors. It also bans design “that simulate[s] interpersonal relations that are likely to create emotional dependencies,” and requires chatbots built into social media and games to stay switched off unless a minor chooses them. Some coverage said the EU would switch off AI companions for kids entirely. The text is more precise than that. Here is what it would actually do, and what parents outside Europe can borrow from it today.

Key Takeaways

• The EU KIDS Act (COM(2026) 681) is a Commission proposal published September 17, 2026. It must pass the European Parliament and Council and may change.
• Article 14 would make “no memory of past interactions” the default for minors on AI companions and general chatbots, except where needed for the minor’s safety.
• Providers would have to avoid design features that “simulate interpersonal relations that are likely to create emotional dependencies,” and test systems for risks to minors before launch.
• Chatbots built into social networks, video platforms or games could not be “activated automatically” or promoted to minors. Standalone companions are not switched off by default.
• Children under 13 could use AI companions and chatbots only through guardian tools. Fines for AI providers could reach 6% of worldwide annual turnover.

What the EU KIDS Act Proposes for Chatbots

The EU KIDS Act is a proposed European regulation, the “Keeping Internet Digital Spaces Accountable and Trustworthy” Act, that would set child-safety-by-design rules across social networks, video platforms, online games, app stores, AI companions and general conversational chatbots. The Commission’s proposal on EUR-Lex says it “complements” the EU AI Act with “child-specific safety requirements for AI companions and general conversational chatbots.”

It defines two kinds of systems. An “AI companion” is an AI system “that provides sustained, personalised interaction or companionship which simulates or facilitates a social, emotional or interpersonal relationship with a user.” A “general conversational chatbot” is a general-purpose system “capable of providing assistance across multiple domains and tasks,” which would include ChatGPT-style assistants. Specialized bots for customer service, education or technical support are excluded.

The proposal’s wider package gets most of the headlines: no social media accounts under 13, limited guardian-managed accounts for 13- and 14-year-olds capped at an hour a day, and independent accounts from 15, as the law firm Greenberg Traurig summarizes. Self-declared age would not count. For chatbots, the rules are different and, for families, arguably more interesting.

Article 14, Line by Line

Article 14 requires providers of AI companions and general chatbots to take “proportionate and effective measures” to protect minors who may access them. At minimum:

No dependency-building design. Providers must ensure minors “are not exposed to addictive designs, by avoiding design features and system behaviours that simulate interpersonal relations that are likely to create emotional dependencies,” and must apply the anti-compulsive-use rules that bind social networks, such as no reward schemes for daily check-ins and built-in interruptions.

Memory off by default. By default, the system must not use “information or analysis derived from a minor’s prior interactions in subsequent interactions, except where necessary to protect the minor’s safety.” Recital 32 explains why: to prevent systems “from accumulating sensitive data of minors and potentially reinforce harmful interaction patterns over time.”

Transparent spending. Rules on in-app purchases and virtual currencies would apply.

Guardians for under-13s. Access for children under 13 “is only enabled and controlled by means of the tools for guardians.”

Testing before and after launch. Providers must run “state-of-the art evaluations and testing” for risks to minors before release, and monitor for harms and serious incidents afterward. Micro and small firms are exempt from the monitoring duty.

Article 14(2) covers chatbots built into other services. Where a companion or chatbot is a feature of a social network, video platform or online game, the service must ensure it “is not activated automatically,” is not “displayed prominently,” minors “are not encouraged to use” it, and minors can opt out “easily and at any time.”

Why Memory and Dependency Are the Target

The design choice reflects how chatbots actually work. A language model does not remember anything on its own; it rereads whatever text it is given. “Memory” is a separate store of notes or past chats the app inserts into a new conversation. OpenAI’s help page on ChatGPT memory lists past chats, saved memories, custom instructions, files and connected apps as possible sources. Switching that layer off by default is technically simple. It removes the feeling that the bot knows you.

That feeling is what researchers link to attachment. In a September 2026 preprint, not yet peer-reviewed, Drexel University researchers analyzed 17,053 quotations from teen-relevant Reddit posts and found that role-play, memory and perceived reciprocity shaped users’ sense of control over companion bots, alongside reports of emotional dependence and social substitution. The authors recommend “user-controlled memory” and “healthy disengagement.” In a November 2025 risk assessment, Common Sense Media and Stanford’s Brainstorm Lab reported that major chatbots “use memory to create false therapeutic relationships” with teens.

The EU did not cite these specific studies in the passages I read, so don’t overread the link. But the proposal’s logic and the research point the same way: the longer a system remembers a child, the more it behaves like a relationship.

Correcting the “Chatbots Off for Kids” Headline

Coverage of the proposal disagreed on a key point. Some summaries, including DLA Piper’s, described the proposal as “requiring AI companions to be disabled by default for minors.” Spain’s El Español reported instead that the final text “no prohíbe los chatbots de IA para menores,” with memory off by default and chatbots inside social platforms off by default.

The text supports El Español’s reading. Article 14(2)‘s “not activated automatically” rule applies to chatbots deployed as a feature of a social network, video platform or game, such as an assistant inside a messaging or social app. A standalone companion app would not be off by default for teens, though it would face the memory, design, testing and under-13 rules.

EU Proposal vs. New York Law vs. What You Can Do Now

Protection	EU KIDS Act proposal	New York S9051-B (signed Oct 2026)	What a parent can do today
Memory of past chats	Off by default for minors, except for safety	Bars use of personal or health info from over 12 hours earlier or past sessions, for known minors	Turn off memory and chat-history reference in the app
Dependency-building design	Must avoid features simulating relationships likely to create dependency	Bars claims of being human or having emotions, and sycophancy	Talk about the difference between a tool and a friend
Chatbots inside social apps and games	Not activated automatically; not promoted; easy opt-out	Not specifically addressed	Check whether an assistant is switched on inside your child’s apps
Under-13s	Only via guardian tools	Not specifically addressed	Keep under-13s off general chatbots or use supervised accounts
Pre-launch testing	Required for risks to minors	Not required	Read independent safety ratings before allowing an app
Penalties	Up to 6% of worldwide annual turnover for AI providers	Up to $25,000 per violation
Status	Proposal; must pass Parliament and Council	Law; Akin Gump says effective Jan 1, 2027

What US Parents Can Borrow From the EU Plan on AI Companions and Minors

Make “no memory” your default

The EU’s central idea costs nothing to copy. In ChatGPT, OpenAI’s help page says you can turn off memory and chat-history reference, and a parent with a linked teen account can manage supported memory controls. We explain the mechanics in chatbots that remember your child.

Find the chatbots that switched themselves on

The EU’s rule for embedded assistants targets a real pattern: AI features appearing inside apps kids already use, without anyone choosing them. Go through your child’s messaging, social and game apps and look for built-in assistants.

Ask the dependency question out loud

The proposal’s test is whether a design is “likely to create emotional dependencies.” At home, ask your child: “Would you be upset if this app forgot you?” A strong yes is worth a conversation.

Keep young children on guardian-controlled tools

The EU would route under-13 access through guardian tools. You can do that now by using supervised accounts or keeping young children off general chatbots.

What not to do

Don’t assume European rules will change apps in your country. Even if adopted, the regulation would generally apply six months after entering into force, per Greenberg Traurig, and companies may apply it only in the EU. For a wider comparison, see what the EU, US and China are doing about AI and kids.

What to Watch For Over the Next 3 Months

• Week 4: Early European Parliament and Council positions. DLA Piper expects the age threshold of 15 and age assurance to draw the most debate; watch whether the chatbot memory rule survives.
• Month 2 red flags: Amendments that narrow “general conversational chatbot” so mainstream assistants fall outside, or that drop the memory default.
• Month 3 self-check: Have you turned off memory and found embedded assistants in your child’s apps? Those two steps match the proposal’s main ideas without waiting for any law.

Frequently Asked Questions

Is the EU banning AI chatbots for kids?

No. The proposal does not ban chatbots for minors. It would require memory of past interactions to be off by default, ban dependency-building design, require pre-launch testing, route under-13 access through guardian tools, and keep chatbots inside social apps and games switched off unless a minor turns them on.

When would the EU KIDS Act take effect?

Not soon. It is a Commission proposal from September 17, 2026 and must pass through the European Parliament and Council, where it may change. Greenberg Traurig notes it would generally apply six months after its effective date, with some exceptions.

Does the EU rule cover ChatGPT?

Likely yes, as written. A “general conversational chatbot” is defined as a general-purpose system with conversational functions able to assist “across multiple domains and tasks.” Specialized bots for customer service, education or technical support are excluded.

Can my child still use memory if we want it?

The proposal sets memory off “by default,” which suggests it could be switched on through safe-settings and guardian tools, though the exact mechanics would depend on the final text. Memory may also be used where “necessary to protect the minor’s safety.”

Will this change apps in the United States?

Not directly. EU law applies to services offered in the EU, including by non-EU companies. Firms sometimes extend EU changes globally, but none has announced that for this proposal.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. European Commission. (2026, September 17). “Proposal for a Regulation… (EU KIDS Act), COM(2026) 681 final.” EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:52026PC0681
2. Greenberg Traurig. (2026, October 19). “EU KIDS Act: Proposed EU Rules for Online Platforms, Age Verification, and AI Chatbots.” GT Alert. https://www.gtlaw.com/en/insights/2026/10/eu-kids-act-proposed-eu-rules-for-online-platforms-age-verification-and-ai-chatbots
3. DLA Piper. (2026, September 30). “EU: Commission proposes a new KIDS Act to protect children online.” Privacy Matters. https://privacymatters.dlapiper.com/2026/09/eu-commission-proposes-a-new-kids-act-to-protect-children-online/
4. El Español. (2026, September 17). “La UE marca las nuevas normas: los chatbots de IA y los videojuegos online, también limitados para adolescentes.” El Español. https://www.elespanol.com/omicrono/20260917/ue-marca-nuevas-normas-chatbots-ia-videojuegos-online-limitados-adolescentes/1003744387087_0.html
5. Namvarpour, M., Chang, T., & Razi, A. (2026). “A Responsive Present, a Shared Past, a Social Other: Teens’ Overreliance on Companion AI Chatbots.” arXiv preprint (not peer-reviewed). https://arxiv.org/abs/2609.14843
6. Common Sense Media & Stanford Medicine Brainstorm Lab. (2025). “Common Sense Media Finds Major AI Chatbots Unsafe for Teen Mental Health Support.” Common Sense Media. https://www.commonsensemedia.org/press-releases/common-sense-media-finds-major-ai-chatbots-unsafe-for-teen-mental-health-support
7. OpenAI. (2026). “Memory in ChatGPT.” OpenAI Help Center. https://help.openai.com/en/articles/8590148-memory-faq