Working draft. Everything on this site is open for public comment and subject to revision.

← Library

Feedback · EU KIDS Act

Enrollment Before Contact

Feedback on the EU KIDS Act proposal, COM(2026) 681 final

Filed 5 November 2026 · Have Your Say reference F33581731 · S. Harrison Knudson, J.D.

A proposal for discussion. Feedback on the proposal for a Regulation on keeping internet digital spaces accountable and trustworthy for children (EU KIDS Act), submitted by the International Foundation for Child-Safe Internet Standards (IFC-SiS). Download PDF

Summary

IFC-SiS supports the KIDS Act. We propose one addition to Article 14, the article governing AI companions and general conversational chatbots: enrollment before conversation.

The rule has two tiers:

  1. Safety duties for every provider. The Article 14 duties apply to every provider whose system minors may access, whether or not any minor is enrolled. Declining to enrol a child never relieves a provider of them.

  2. Enrollment for conversation. A provider shall not let its system converse with a minor unless a person holding parental responsibility has enrolled that minor for a defined role, such as tutoring in a subject or help with homework.

Three principles follow from it:

Enrollment is a narrow, checkable rule. It gives the Special Panel's call for providers to carry the burden of proof something concrete to prove.

What Article 14 gets right, and the gap

Article 14 already holds the two halves of the rule we propose. It needs one connection between them.

What it gets right. It places safety-by-design duties on every provider of an AI companion or general conversational chatbot that minors may access: no design that fosters emotional dependency, no carrying a child's earlier conversations into later ones by default, testing before market and monitoring after. It also recognises that a guardian should stand between a young child and a chatbot: under-13s may use these systems only through tools for guardians.

The gap. That guardian step stops at 13. From 13 to 17, a commercial AI system may form a sustained, personalised relationship with a child that no guardian has authorised. By the Act's own definition, sustained personalised interaction is what an AI companion provides.

A second gap: coverage. Article 3(5) excludes from "general conversational chatbot" systems whose conversation is limited to a specialised service. A tutoring or homework system that is not an AI companion could fall outside Article 14 altogether, though it may hold the most sustained relationship with a child of any AI system. Amendment 2 closes this by function, not label.

The Act already stops unknown users from messaging children on social networks: children do not appear in contact suggestions and cannot be added to groups without agreement. Those rules govern other users reaching a child. On a chatbot, the provider's own system is the one reaching the child, so no contact rule applies. The AI is the contact.

Offline, the institutions entrusted with a child's education, care and development work within recognised frameworks of authorisation, professional responsibility and safeguarding. A school, a doctor, a tutor, a coach and a camp ordinarily begin with a parent's consent. The internet is the one place a stranger can form a relationship with a child without that step. Enrollment restores it.

Why guardian tools are not enough. A tool for guardians lets a parent supervise a relationship that has already begun. Enrollment decides whether it begins. Recital 43 rightly says compliance should not depend exclusively on guardians, and tools that parents must find, configure and maintain on every device do depend on them. One study of teenagers' AI use found only 32% of parents feel sufficiently supported, and 42% receive no support at all. Enrollment asks a parent for one act, once, and places every continuing duty on the provider.

Enrollment asks the Union to apply online a principle it and every Member State already recognise offline: the child holds the rights, and the parent is the recognised agent for exercising them.

Controls that evolve with the child. Enrollment is not a rule that parents decide everything. The terms a guardian sets for a 9-year-old and a 16-year-old will differ, as Article 5's evolving-capacities language requires. As a child matures, the guardian's terms widen. A teenager must also be able to reach help without a guardian: emergency services, child helplines, and confidential health, counselling or child-protection services. Amendment 6 keeps those routes open as Union and national law define them; a provider does not open or close them on its own judgement.

Proposed amendments

Eight amendments carry the rule. They add to the Commission's text and remove nothing except the under-13 limit on the guardian step.

Amendment 1: Article 3, new definitions

(x) 'conversation' means output that an AI system generates for and directs to a user in an exchange. It does not include a list of search results, autocomplete, spelling or grammar correction, or the display of content the system did not generate for that user. Whether an interaction is a conversation depends on its function, not on how the provider describes the service;

(y) 'enrollment' means the affirmative, verified authorisation of a guardian, given to a provider, for a specific minor to converse with an AI system in a defined role, recorded by a child-specific authorisation credential that does not require disclosure of the minor's legal identity;

(z) 'defined role' means the purpose and limits of conversation set by the guardian at enrollment, such as tutoring in a subject or help with homework, taking account of the minor's age, maturity and evolving capacities.

Amendment 2: Article 14, new paragraph 1e (coverage)

Article 3(5) excludes from "general conversational chatbot" systems whose conversation is limited to a specialised service. As drafted, a tutoring or homework system could fall outside Article 14 altogether, and providers could reclassify to avoid it.

1e. Paragraph 1 and paragraphs 1a to 1d and 1f shall apply, as appropriate, to providers of AI systems accessible to minors that are designed or reasonably capable of sustaining personalised interaction with a user across conversations, including systems providing specialised services such as tutoring or homework assistance, regardless of how the provider designates or classifies them. Any adaptation of those obligations to such specialised systems shall preserve the substantive protections of this Article and be proportionate to the nature, purpose and risks of the system.

Amendment 3: Article 14(1), point (d), replaced

(d) the system does not engage a minor in conversation except in accordance with paragraph 1a; for minors under 13, enrollment shall be given and managed through tools for guardians.

Amendment 4: Article 14, new paragraph 1a (enrollment before conversation)

1a. Where a provider is aware, or on the basis of objective indications should be aware, that a user is a minor, it shall not allow its system to engage that minor in conversation unless a guardian has enrolled the minor for a defined role, subject to paragraphs 1d and 1f. In particular:

(a) conversation shall remain within the defined role. Where a minor asks for something outside it, the system shall decline to widen the role, may point the minor to suitable help, and shall inform the guardian where the terms of enrollment so provide, unless doing so could expose the minor to a reasonably foreseeable risk of harm or would disclose a request for help protected by paragraph 1d. The role shall not be widened without new enrollment;

(b) the guardian may narrow, modify or revoke enrollment at any time. On revocation the provider shall delete the minor's conversation records and any stored profile or memory of the minor, except data that Union or national law requires it to keep or that is strictly necessary to preserve evidence of a safety incident, which shall be kept under restricted access and used for no other purpose;

(c) enrollment shall be verified in accordance with Article 26, by methods that are proportionate, use no more data than verification requires, do not depend exclusively on government-issued identification, do not locate, track or profile adults or minors, and do not require disclosure of the minor's legal identity where a child-specific credential suffices;

(d) an educational establishment may enrol a pupil only for a defined educational role and only with the consent of the pupil's guardian;

(e) a provider that implements enrollment verification in accordance with this Regulation, tests its effectiveness regularly and corrects vulnerabilities known to it shall not be regarded as infringing this paragraph solely because a minor circumvents it, unless the provider was aware or should have been aware of the circumvention;

(f) enrollment shall take due account of the minor's views, age and maturity, and shall not authorise a provider to require a minor to take part in a conversation.

Amendment 5: Article 14, new paragraph 1b (enrollment is additional)

1b. Enrollment is additional to, and never a substitute for, the obligations in paragraph 1 and elsewhere in this Regulation. Those obligations apply to every provider within the scope of this Article whose system minors may access, whether or not any minor is enrolled, and a provider's failure or refusal to enrol a minor shall not relieve it of them.

Amendment 6: Article 14, new paragraphs 1c, 1d and 1f (role, confidential help, disclosures)

1c. Providers shall maintain safeguards appropriate to the defined role, informed by the established norms of the profession that ordinarily performs that role with children. Performing a role does not confer any professional qualification or authorise any act that Union or national law reserves to a regulated profession.

1d. Nothing in paragraph 1a shall prevent a system from directing a minor to emergency services, child helplines, or confidential health, counselling or child-protection services that Union or national law allows a minor to reach without a guardian, or require the guardian to be told of that referral. Providers shall establish procedures for responding to credible disclosures of abuse or imminent danger that come to their attention through a conversation, including referral to competent authorities where Union or national law requires it. Such a disclosure shall not be referred to the guardian where that could endanger the minor. This paragraph imposes no general obligation to monitor conversations.

1f. Paragraph 1a does not apply to an AI system operated by or on behalf of a child helpline, or a confidential health, counselling or child-protection service, where Union or national law allows minors to reach that service without a guardian. Such systems remain subject to paragraph 1 and to the safeguarding duties of the service that operates them.

Amendment 7: Article 26(1), enrollment added

Article 26(1) names the purposes for which parental responsibility is verified. Enrollment is not yet among them.

Article 26(1): add "enrollment under Article 14(1a)" to the purposes listed, so that the free electronic means Member States provide under Article 31(1) is also available for enrollment.

Amendment 8: Article 14(3), codes of conduct and standards

Two routes with different legal effect. The code route exists in the Commission's text and gives an indication of compliance. The presumption of conformity is a new proposal of this submission and a stronger consequence; it does not follow from Regulation (EU) No 1025/2012 unless the co-legislators write it into the KIDS Act:

Codes of conduct. A code of conduct under Article 23 may set out how providers meet paragraphs 1a to 1d, including role-based safeguards drawn from professional norms. Adherence to a code the Commission has assessed as adequate may serve as an indication of compliance.

Harmonised standards. The Commission shall request one or more European standardisation organisations, in accordance with Article 10 of Regulation (EU) No 1025/2012, to draft European standards on enrollment, the defined role and role-based safeguards. Systems that conform to harmonised standards, or parts thereof, the references of which have been published in the Official Journal of the European Union, shall be presumed to conform with paragraphs 1a to 1d to the extent those standards cover them.

A recital to accompany them. Age assurance under Articles 27 to 29 continues to apply as the Regulation provides; the enrollment gate applies once a provider is aware, or should be aware, that a user is a minor, and the safety duties in paragraph 1 apply to every user who may be a minor. The amendments impose no general restriction on access to information: search, reading and public spaces are untouched, and paragraph 1d protects a minor's independent access to confidential help. They require no data beyond what enrollment and verification need, and no monitoring of conversations beyond paragraph 1d.

Why it fits the Act, and the objections

Enrollment is built from parts the Act already contains: guardian verification, guardian tools, safety-by-design duties and codes of conduct. It adds the order in which they apply.

It makes the burden of proof provable. The Special Panel recommended shifting the burden of proof to providers, so that they demonstrate their services are safe for minors. "Was this child enrolled, and did the system stay within its role?" is a question of fact a regulator can answer from records. "Is this product safe for children?" is a judgement no record settles.

It reuses rules that already exist. The professions that work with children already have norms for conduct, scope and reporting abuse. An AI system enrolled to tutor is measured against what is expected of a tutor. The co-legislators need not write those norms; they can reference them, as Regulation (EU) No 1025/2012 already provides for European standards.

The strongest objections, and the answers:

An offer to test it. IFC-SiS would welcome an independently evaluated pilot with one or more Member States, for example through enrolment records that schools already hold, measuring uptake, error rates and burden before the rule applies at scale.

Sources

S. Harrison Knudson, J.D., Director, International Foundation for Child-Safe Internet Standards (IFC-SiS) · shknudson@ifcsis.org · 202-684-6309

This feedback is a proposal meant for discussion. Critical reactions — doctrinal, comparative or practical — are more useful than supportive ones.

Send a comment