AI chatbots & AI companions

Standalone and embedded AI

The two definitions can cover standalone systems. Article 14(2) also expressly regulates AI companions or general conversational chatbots deployed as a feature inside a social network, video-sharing service or online game.

For AI embedded in a social network or video-sharing service, Article 8(3) limits which parts of Chapter III apply to that AI functionality: the AI-specific rules, age-appropriate access rules and general agency/tool rules. The host service’s own obligations still apply in their proper scope. An online game remains subject to Article 15, while its embedded qualifying AI also engages Article 14(2).

Emotional dependency and compulsive use

Providers must use proportionate, effective measures to protect minors’ health, safety, fundamental rights, well-being and development. Article 14(1)(a) requires them to avoid design features and system behaviour that simulate interpersonal relationships likely to create emotional dependencies.

The provider must also apply Article 9’s general prohibition on design reasonably foreseeable to encourage compulsive or excessive use, and put in place time-limiting and interruption measures that protect school time and core sleep hours.

Safe settings and previous-interaction memory

Article 14 applies Article 11’s safe-setting approach. By default, the system must not use information or analysis from a child’s previous interactions in later interactions, except where necessary to protect the child or give effect to the required safety settings.

This does not say an AI system may never retain any information. It creates a default rule against reusing prior-interaction information for subsequent exchanges, with the specified safety exceptions.

Economic transactions

Article 14 applies Article 13 to the two AI categories. A child must be clearly told in real time when a transaction is economic, and purchasable virtual currency must display the corresponding official-currency value. The system must not be designed to lead to excessive, impulsive or unwanted spending, including exposure to the variable-reward systems described in Article 13.

Access below age 13

Access by a child below 13 must be enabled and controlled only through Article 20 guardian tools. This is an AI-specific condition in Article 14(1)(d), not the same as Article 7’s guardian-account regime for child-designed video-sharing services.

Guardian tools must be age-tailored, accessible, secure against easy circumvention and respectful of the child’s privacy and agency. The child must know when they are active.

Pre-market testing and post-market monitoring

Before placing the system on the market or putting it into service, the provider must perform state-of-the-art evaluation and testing for likely risks to minors’ health, safety, fundamental rights, and physical, mental and emotional well-being and development. Appropriate safeguards must address the identified risks.

Providers must then monitor after market entry to identify, assess and, where appropriate, mitigate harm and emerging risks, including serious incidents involving minors.

Controls for embedded AI

Where qualifying AI is embedded in a social network, video-sharing service or online game, the host provider must ensure that:

• the AI is not activated automatically;
• the feature is not displayed prominently;
• minors are not encouraged to use it; and
• once enabled, minors can opt out easily and at any time.

These rules address placement and choice in the host service; they do not convert every embedded automated feature into an AI companion or general conversational chatbot.

Agency, reporting and guardian support

Articles 18–20 add understandable controls, child-friendly reporting and guardian tools. Minors must be able to control and provide feedback on prompts and suggested information. Reports must be prioritised and followed by support and redress information.

Article 21 also provides complaint and representation routes for users of these AI systems. Providers outside the EU that offer covered systems in the EU must address the Article 24 legal-representative rule.

Age assurance

Because Article 8 applies to the two AI categories, providers may depart from child-safe defaults only after compliant age assurance establishes that the user is an adult. Article 29(4) permits methods other than the certified EU Article 6 verification route if the provider can demonstrate compliance with Articles 27 and 28.

See age assurance & age verification.

Relationship with the AI Act

The proposal imports the AI Act definitions of “AI system”, “general-purpose AI system” and “provider”. It says the KIDS Act rules would complement, and operate without prejudice to, the AI Act.

Article 34 uses the AI Act’s Chapter IX supervisory structure for the two AI categories. Intentional or negligent non-compliance with the relevant KIDS Act obligations could attract an administrative fine under AI Act Article 99 of up to 6% of the provider undertaking’s worldwide annual turnover. That is a category- and infringement-specific route, not a universal penalty for every provider. See compliance, enforcement and implementation for the authority and procedure map.