AI Companion Apps Under Scrutiny: Age Limits, Regulatory Risks and How to Lock Them Down

AI companion and “virtual friend” applications are not built for minors, and the most prominent services now say so explicitly. Replika requires users to be 18 or older and carries an 18+ rating on the Spanish App Store. Character.AI stripped open-ended character chat from users under 18 in late 2025, and Meta announced in January 2026 that it was pausing teenager access to its AI characters worldwide. What still does not exist is a specific European legal age for these services.

  • Age gates vary wildly: Replika enforces 18+, Character.AI declares 13+ but has removed open chat for minors, and Meta has paused teen access entirely.
  • Regulators agree on the core risk: the FTC, Australia’s eSafety Commissioner and Italy’s Garante all point to engagement-driven design and missing age verification, not just harmful chatbot responses.
  • EU law stops short of an age limit: the AI Act bans exploiting age-based vulnerabilities and requires chatbot disclosure, but a 16-year minimum remains only a non-binding parliamentary resolution.

What age do companion chatbots actually require?

In practice, the answer for the best-known services is 18. Replika, operated by Luka Inc., sets that threshold according to the company itself, a position documented by Australia’s eSafety Commissioner, and the Spanish App Store classifies it as 18+.

Character.AI states a minimum age of 13, but on 29 October 2025 it announced it would remove open-ended AI conversation from users under 18. According to its 21 November update, the rollout began on 24 November in the United States and continued in other markets afterward.

Meta, for its part, paused adolescent access to its AI characters globally, a measure announced on 23 January 2026. The company describes the pause as temporary while it builds a new version with parental controls; the general Meta AI assistant remains available to teens.

General-purpose assistants such as ChatGPT, Gemini and Claude follow their own separate rules, distinct from products designed to converse with you every day.

The risks regulators keep flagging

Three concerns recur across official bodies, and none of them concerns a single harmful message. The first is that a minor comes to trust the chatbot like a person. The second is that the design pushes compulsive use. The third is the absence of real age verification.

The US Federal Trade Commission sent information requests to seven companies on 11 September 2025, including Alphabet, Character Technologies, Meta and OpenAI. Its argument is that these chatbots effectively imitate human traits, emotions and intentions, and that by presenting themselves as friends or confidants they can lead children and adolescents to trust them and form relationships.

Australia’s eSafety Commissioner went further. Its safety advisory warns that these services are often designed to encourage continuous interaction, which can become addictive. In a later report it reviewed Character.AI, Nomi, Chai and Chub AI and concluded that none had age controls worthy of the name. Its survey of 1,950 minors aged 10 to 17 found that 79% had used an AI companion or assistant, and 8% had used a companion specifically.

In Europe, Italy’s Garante fined Luka Inc., Replika’s parent company, 5 million euros on 19 May 2025, citing among other things the absence of age verification and judging its system at the time still deficient. The company has appealed the sanction, according to the Garante itself.

The underlying risk lies in combining a system that agrees with you and a business that needs you to come back tomorrow.

That first trait is a general feature of language models, which tend to be agreeable; in an app whose business is companionship, that flaw stops being a defect and becomes the product.

What European and Spanish law requires today

The European AI Act sets no age, but it imposes two limits that directly affect these apps. Since 2 February 2025, it has been prohibited to use AI systems that exploit age-related vulnerabilities to materially distort someone’s behaviour and cause them significant harm. And since 2 August 2026, Article 50 requires that anyone talking to a chatbot knows it is an AI, according to the Commission, from the start of the first interaction unless that is obvious.

The European Parliament approved a report on 26 November 2025 by 483 votes in favour, 92 against and 86 abstentions, calling for a digital minimum age of 16 for social networks, video platforms and AI companions, with access between 13 and 16 allowed if parents authorise it. It is a non-legislative resolution: it binds no one.

In Spain, the legal reference remains Article 7 of the Organic Law on Data Protection: a minor can only consent to the processing of their data from the age of 14.

How to configure or block these apps

The most effective tool sits on the minor’s phone, not in the app: if the store filters by age, an app rated 18+ cannot be installed.

  • iPhone: in Screen Time, content restrictions let you limit apps by age rating and prevent the minor from installing applications without permission, according to INCIBE.
  • Android: in Family Link, Controls > Google Play lets you choose the maximum content rating that can be downloaded, according to Google Play’s help pages.
  • The browser is the back door: Character.AI also works from its website, so a store filter is not enough. A web filter is needed, or better still, a conversation.

Adults using these apps should review what is stored about them. Character.AI allows users to export their data or delete their account from the app or the web, and doing so soon is advisable: the company filed for Chapter 11 bankruptcy protection on 12 October 2026 to sell the company under court supervision, and warns that if no offer allows it to keep operating, the platform will close. According to its court notice of 5 November, Disney is the reference bid for part of its assets; as of 17 November 2026, the auction is scheduled for 1 December.

If a minor is struggling with one of these relationships, INCIBE’s Cybersecurity Help Line, 017, is free and confidential.

Self-regulation has arrived late and under pressure. Character.AI cut off minors after receiving regulator questions, Meta paused its characters to rebuild them, and Replika was fined partly for failing to verify ages. That the specific European obligation for these chatbots is, for now, to disclose that they are an AI says much about how little thought went into a product whose promise is precisely that you forget it. The 16-year threshold the European Parliament wants will likely arrive, and age will be the least of it. The decisive step will be limiting designs that reward dependence, and no rule yet specifies how to measure that.

Sources & References