The proposal reaches inside the conversation

The European Commission proposed new rules on 17 September that would require AI companions and general conversational chatbots to change how they interact with children. The KIDS Act pairs social-media age restrictions with requirements for the design of services that minors can use.

For AI products, the significant detail is not simply an age gate. Article 14 of the draft addresses features and behaviours that imitate interpersonal relationships in ways likely to create emotional dependencies. It also proposes that information from a minor's earlier interactions should not be carried into later ones by default, with exceptions for safety and the required protective settings.

Those are proposed obligations. The Commission has adopted its proposal, not enacted a final EU law.

Different services, different access rules

The Commission's account of the social-media rules describes no accounts below 13, guardian-supervised accounts at 13 and 14, and independent accounts from 15. The age provisions target the social-networking and video-sharing services in scope.

Chatbots are addressed separately. The draft would allow access below 13 only through tools controlled by a guardian. Calling the package a ban on all AI for under-15s would therefore misstate it.

The definitions matter too. A general conversational chatbot is defined as a system providing assistance across multiple domains. The draft excludes conversational functions limited to a specialised service or predefined set of tasks, such as technical support. The label on a product alone would not explain which provision applies.

A launch test would not be the whole job

The draft calls for testing risks to children before a covered AI system reaches users and for appropriate safeguards. It also provides for monitoring emerging harms afterwards, although the monitoring clause contains an exception for micro and small enterprises.

When a chatbot is built into a social platform, video service or online game, the proposed rules would stop automatic activation and prominent placement. Children would not be encouraged to use it and would need an easy way to leave it.

The broader package includes safer recommendation systems and account defaults. Its age-assurance provisions seek to establish whether a threshold is met without turning the check into a new means of tracking someone. Those privacy protections are requirements in a proposal, not evidence that every implementation already meets them.

What to watch next

For product teams, the proposal makes a useful distinction: deciding who may enter a service is different from deciding what that service does once a child is inside. A birthday check does not address memory, repeated prompts or the impression of a personal bond.

For readers, the next questions are the final scope, the evidence required to demonstrate compliance and how enforcement works across different products. None is settled by an announcement headline.

The draft sets application periods relative to entry into force. It does not establish a fixed date on which these new chatbot requirements become binding.

Sources

  1. European Commission: KIDS Act announcement17 September 2026 announcement of the proposed child-safety package. Adoption by the Commission is not final legislative enactment.
  2. European Commission: proposed regulation and supporting documentsOriginal proposal downloaded and read: definitions in Article 3, chatbot obligations in Article 14, age-assurance safeguards in Articles 27–29 and application provisions in Article 43.
  3. European Commission: KIDS Act questions and answersPublic explanation of the account-age stages, parental controls and the difference between access rules and safety-by-design obligations.
  4. European Commission: KIDS Act overviewScope, recommendation systems, account settings, proposed exclusions and enforcement outline.