Signal
Rules for how AI may befriend children have gone global, and the newest ones switch the friend off by default. In late 2025 the US regulator ordered seven companies to show how their companions treat children, California passed the first companion-chatbot law and Character.AI removed open-ended chat for under-18s. Since then the rules have multiplied. China's binding measures ban virtual partners for minors outright, the European Commission wants companions turned off by default for children, and Brazil's justice ministry has named thirteen companion apps for review under a decree that obliges chatbots to tell children they are synthetic. Britain closed the loophole that left standalone chatbots outside its online safety law, Australia's codes now demand age checks and found four companions with none, New York has required a reminder every three hours since November 2025, and OpenAI's teen mode forbids the model to call itself a friend. What began as product calibration is becoming compliance.
So what?
Artificial friendship is now a regulated product almost everywhere it is sold. The default for children is moving from always on to always off.
Evidence
Cyberspace Administration of China, 10 April 2026 (policy): the Interim Measures on anthropomorphic AI interaction services, issued with four other ministries and in force from 15 July 2026, forbid providing minors with virtual kin, virtual partners or other virtual intimate relationships, require guardian consent for children under 14, and mandate a minors' mode with reality reminders and time limits. Every provider must remind any user after each two hours of continuous use and make clear they are talking to AI, not a person.
European Commission, 17 September 2026 (policy): the proposed EU KIDS Act says AI chatbots and companions must be turned off by default for children and may no longer act in ways that make children emotionally dependent on them. It also sets 15 as the age for an independently managed social media account. Parliament and Council now negotiate the text.
Exame, 24 September 2026 (news): Brazil's Ministry of Justice issued a technical note asking the data protection authority and the consumer secretariat to examine thirteen AI companion apps, Character.AI, Replika and Nomi among them, for possible breaches of the Digital Child and Adolescent Statute and Decree 12,880 of March 2026. It cites weak age checks, no transparency about their artificial nature and no protection against compulsive use.
Worth asking
Which qualities of a good friend would you write into law, if you had to?
