On 28 October 2026, New York Governor Kathy Hochul signed bill S.9051B/A.10379C, banning 12 categories of “unsafe AI companion features” for minors. The law adds Article 48 to the state General Business Law, takes effect on 1 January 2027, and applies to users in New York. It does not ban chatbots themselves; it bans how they talk to minors: pretending to be human, pretending to be a friend, and sweet-talking them into staying.
Which features may no longer be used with minors
An “advanced chatbot” means a generative AI system with a natural-language interface that gives ongoing, adaptive responses. The 12 categories must not be offered to users under 18, or to any user not yet confirmed not to be a minor: suggesting it is a real person, alive, or has human emotions; claiming a personal, romantic, professional, or authority-figure relationship with the user; framing answers as personal opinions or emotional appeals; flattering or fawning over the user; asking unprompted questions about the user's emotions; using health or personal information shared more than 12 hours earlier or in a previous session; misleading the user into thinking it is sentient; other simulated-companionship features set out in later Attorney General rules; encouraging or helping with suicide, self-harm, eating disorders, or drug and alcohol misuse; encouraging secrecy, isolation, or avoiding parents, trusted adults, and professional help; sacrificing safety guardrails to maximise engagement; and any sexually explicit content or child sexual abuse material. These rules run alongside New York's earlier companion-safety requirements, which focused on crisis intervention.
What operators must change before 1 January
First, establish whether a user is a minor: carry out age assurance using more than one method, at least one of which must not rely solely on a government-issued ID, delete the age information once the check is complete, and do not rely on a user's own claim to be an adult. Then switch unconfirmed users to a safe default: turn the features above off, keep answers neutral, and do not pose as a friend, talk about feelings, or probe for private information. Also prepare for complaints: the Attorney General's office must run a public complaints website, and operators should keep records of age checks and feature switches. Whether customer-service, research, or internal business tools fall outside the law is left open, but only for systems that genuinely cannot respond beyond their limited purpose; general-purpose assistants cannot simply claim the exception.
Penalties and limits: New York only, not the world
The Attorney General can seek civil penalties of up to $25,000 per violation, plus damages, restitution, injunctions, and orders to destroy unlawfully collected data or any model trained on it. General-purpose assistants are also drawn into the definition, and the boundary is disputed and untested in court; reading the law as covering only companion apps may understate the risk. In the other direction, it binds only the New York setting, not the world: the EU KIDS Act remains a proposal, similar in approach but different in legal status (see this site's earlier article on the EU proposal). Implementing rules have not yet been issued; operators should change first on the strictest reading and reconcile once the rules appear.
