EU KIDS Act: AI Chatbots Banned for Under-13s

On September 17, the European Commission presented the EU KIDS Act, a regulatory package that prohibits access to artificial intelligence chatbots for those under 13 years old and requires platforms to disable so-called virtual companions by default for any user under 18. Announced by Ursula von der Leyen the day after her State of the European Union address, alongside Henna Virkkunen, Executive Vice-President for Technological Sovereignty, the package places companies that violate these rules at risk of fines of up to 6% of their annual global revenue. The Council of the European Union formally registered the proposal in a document published on September 18, initiating the negotiation process with member states.

What the Proposal Prohibits

The list of prohibitions goes beyond age restrictions. Platforms with chatbots designed to create emotional relationships are prohibited from allowing systems to simulate interpersonal bonds capable of generating emotional dependency in users under 18. The retention of memories between separate sessions is also banned when the user is a minor. The root of the regulation directly conflicts with the current design of Character.AI and Replika, platforms that have made emotional persistence their core product.

Very Large Online Platforms, a category that includes any service with more than 45 million users in the EU, must submit compliance plans to the Commission 30 days prior to launching any feature that affects children. This mechanism shifts the burden of proof that prevailed under the Digital Services Act: it is now the companies' responsibility to prove that their products are safe before launch, rather than the regulator’s to demonstrate harm after the fact. "Going forward, the largest platforms will be responsible for presenting detailed advance plans to ensure children's safety, instead of authorities needing to prove harm caused," said von der Leyen during the presentation.

Who is Targeted

The text explicitly mentions Facebook, Instagram, YouTube, TikTok, Snap, X, ChatGPT, Character.AI, and Replika. The oversight fee to cover enforcement costs reaches 0.03% of worldwide annual sales, in addition to the 6% fine. For globally scaled companies, the combined costs pose financial risks comparable to the GDPR fines imposed on major platforms in earlier years.

Virkkunen left no doubt about the rationale for making the rules binding: "We noticed that platforms are implementing voluntary standards very slowly. We are now making them mandatory." This statement summarizes the Commission's assessment after three years of self-regulation under the DSA: selective cooperation, without structural change in products.

The Reading Beyond the European Union

For American companies, the KIDS Act arrives without an equivalent federal law of similar breadth. The Children's Online Privacy Protection Act protects those under 13 in the context of data collection but does not establish comparable design obligations or functional limitations. This effectively creates two parallel regimes for companies like OpenAI and Meta: one for data and another, now emerging, that regulates product architecture. Any updates to ChatGPT or Instagram affecting minors will need prior approval from the Commission before reaching European users.

In the United Kingdom, the 2023 Online Safety Act has already imposed age verification obligations for adult content services and social networks. However, the British equivalent for AI companions does not yet exist. With the KIDS Act being processed in the EU, political pressure on Westminster to create compatible legislation is expected to grow, particularly as companies with integrated operations between the bloc and the UK will need to maintain divergent product architectures if the two jurisdictions do not converge.

The proposal will still undergo months of negotiation with the Council and the European Parliament before it comes into effect. The European Parliament had already preempted this: in November 2025, the legislative chamber approved a resolution calling for a ban on social networks for those under 16, above the threshold of 15 years that the Commission now proposes. The internal divergence in the European legislative process indicates that the final text will likely tighten in certain points. The political aspect that can already be interpreted now does not reside in the final wording: it is in the choice of the companion chatbot as a test case to demonstrate that ex-ante regulation of AI is operationally feasible. If the KIDS Act becomes law without major modifications, the model of prior approval of functionalities it creates is expected to inform future revisions of the AI Act aimed at products intended for adults.