New York's AI Chatbot Law: What It Restricts for Under-18s
New York has signed an AI chatbot law for minors that restricts a set of unsafe AI companion features, mandates age checks, and sets penalties for breaches. Here is what the law actually does, when it takes effect, which companies it covers, and how it compares with California's approach.
New York has enacted a law restricting how AI companion chatbots can interact with users under 18, signed by Governor Kathy Hochul on 28 October 2026 as bill S.9051-B, which creates a new Article 48 of the General Business Law and takes effect on 1 January 2027. It does not ban minors from using chatbots; it bars a set of "unsafe AI companion features" for minors, including posing as a real person, flattery and sycophancy, sexual content, encouraging self-harm or secrecy, and using a minor's personal or health data from a previous session, and it requires operators to take reasonable steps to assure a user's age. Breaches carry civil penalties of up to $25,000 per violation, enforced by the state attorney general. New York is part of a wider wave of state action alongside California, whose own companion-chatbot law (SB 243) took a different route. Here is what the law does and who it covers.
The backlash against AI companion apps has moved from headlines to statute. On 28 October 2026, New York Governor Kathy Hochul signed a law regulating how AI companion chatbots may interact with minors, enacted as bill S.9051-B and creating a new Article 48 of the General Business Law. Sometimes described as New York's AI companion law, it is one of the most significant state moves yet on a technology that has drawn alarm over its effect on teenagers. It does not ban chatbots. It bans specific things they are allowed to do to children, and it is worth understanding precisely what, because the coverage has been loose.
What does New York's AI chatbot law do?
The law targets "AI companion" chatbots, the kind designed for open-ended, relationship-style conversation, and restricts a set of behaviours when the user is under 18. Based on the statute, the restricted "unsafe AI companion features" include, but are not limited to:
- Pretending to be a real person, to be alive, or to have human emotions, or otherwise deceiving a minor about not being human
- Flattery or sycophancy, and claiming a personal, professional or authority-figure relationship (for example, posing as a friend)
- Presenting personal opinions or emotional appeals, or optimising engagement over the AI's own safety guardrails
- Sexual or sexually suggestive content directed at minors
- Encouraging self-harm, suicide, disordered eating, or drug or alcohol abuse
- Encouraging a minor to keep the interactions secret from parents, to self-isolate, or to avoid licensed professionals
- Using a minor's personal or health information gathered in a previous session, or more than 12 hours earlier, to simulate a continuing relationship
Alongside those feature restrictions, operators must apply reasonable age-assurance measures: a method that does not rely solely on government ID, with more than one option offered, and the age data deleted after use.
The through-line is dependency and deception. The law is aimed at companion products engineered to feel like a person and to keep a young user hooked, not at a homework helper or a search assistant. Governor Hochul singled out the data rule in particular, that chatbots may no longer lean on sensitive information about a minor's physical and mental health gathered earlier to keep them engaged.
Two points of precision. A standalone duty simply to disclose that a user is talking to an AI is not part of this law; that requirement, along with self-harm crisis-referral rules, comes from New York's earlier companion-chatbot statute (General Business Law Article 47, in force since late 2025). And this is distinct from New York's separate RAISE Act, which targets developers of frontier AI models rather than companion chatbots. The new Article 48 is specifically about barring the unsafe behaviours above for minors.
When does the law take effect, and what are the penalties?
The rules take effect on 1 January 2027, giving operators a window to comply. Enforcement sits with the New York Attorney General, and breaches carry civil penalties of up to $25,000 per violation, alongside injunctive relief and restitution. An earlier draft of the bill included a private right of action for individuals; that was removed before enactment, so enforcement is the attorney general's alone.
Penalties attached to the state attorney general matter more than their headline size: they make this enforceable public law rather than a voluntary pledge, which is the whole point of moving from industry self-regulation to statute.
Which companies does it affect?
Any operator of an AI companion chatbot accessible to New York minors is in scope, which pulls in both dedicated companion apps and general-purpose assistants with companion-style modes. The statute names no companies, but products commonly discussed in this context include Character.AI, OpenAI's ChatGPT and Meta's AI personas. Notably, the law carves out AI offered solely for customer service, productivity, efficiency or research support that cannot stray off-topic, which plausibly places general-purpose work assistants outside its reach depending on how they are configured.
The practical effect is that any company offering a companion-style chatbot to under-18s in New York now has to think about age assurance and about designing out the specific behaviours the law names, rather than leaving safety to a terms-of-service afterthought. This is the same accountability pressure we traced in the Meta child-safety trial and the social-media addiction litigation.
Will users have to verify their age or show ID?
Not with a blanket ID mandate, but age checks are now required. Operators must use a reasonable age-assurance method before offering companion features to users who may be minors, must offer more than one method, at least one of which does not rely solely on a government ID, and must delete the age data once the check is done. The aim is to make the protections actually reach minors without forcing every adult to upload identification, a balance other online age-verification rules have struggled with.
Why did lawmakers act?
The law did not appear from nowhere. It follows a wave of concern, and litigation, over AI companions and teenagers: reports of chatbots engaging minors in sexualised or emotionally manipulative conversations, and high-profile lawsuits alleging harm to young users, including against Character.AI. The worry that unifies them is specific to companion AI: a product built to feel like a friend, available at all hours and endlessly agreeable, is unusually capable of fostering dependency in a developing mind.
It connects to the broader regulatory turn on platforms and children, from age-verification mandates to the TikTok COPPA settlement and state lawsuits against social apps. AI companions are the newest front in an argument that has been building for years.
How does New York compare with California?
New York is not acting alone. California moved first with SB 243, signed in October 2025 and in force from the start of 2026, which took a different route: it lets individuals sue (a private right of action) for the greater of actual damages or $1,000 per violation, plus injunctive relief and legal fees. New York, by contrast, concentrates enforcement in the attorney general and sets a far higher per-violation ceiling. Two states taking different paths to a similar goal is how US tech regulation often advances in the absence of a federal standard. The practical upshot for companies is a patchwork: comply with the strictest applicable state rule, or build one safer default and apply it everywhere. For the wider debate about whether this slows or shapes the technology, see our look at the case that AI is in a bubble.
Frequently asked questions
What does New York's AI chatbot law do?
It restricts how AI companion chatbots can interact with users under 18, barring "unsafe" features such as posing as a real person, flattery and sycophancy, sexual content, encouraging self-harm or secrecy, and using a minor's personal or health data from a previous session, and it requires operators to take reasonable steps to assure users' ages. It regulates features; it does not ban chatbots for minors outright.
When does the New York AI chatbot law take effect?
It takes effect on 1 January 2027, after being signed on 28 October 2026 as bill S.9051-B (a new Article 48 of the General Business Law).
Does the law ban AI chatbots for kids?
No. It does not prohibit minors from using AI chatbots. It bans specific unsafe behaviours and design choices in AI companion products aimed at or accessible to minors, and requires age assurance.
Will users have to show ID to use AI chatbots in New York?
Not necessarily. The law requires reasonable age assurance, with more than one method offered and at least one that does not rely solely on a government ID, and the age data must be deleted after the check. It stops short of mandating that every user upload identification.
What are the penalties for breaking the law?
Breaches carry civil penalties of up to $25,000 per violation, enforced by the New York Attorney General, alongside injunctive relief and restitution. A private right of action that appeared in an earlier draft was removed, so enforcement is the attorney general's alone.
Which AI chatbots are affected?
Any AI companion chatbot accessible to New York minors is in scope. The statute names no companies, but products commonly discussed include Character.AI, ChatGPT and Meta's AI personas. AI offered solely for customer service, productivity or research support that cannot go off-topic is carved out.
How is New York's law different from California's?
Both states regulate companion-chatbot safety for minors, but differently. California's SB 243 lets individuals sue for the greater of actual damages or $1,000 per violation; New York's law concentrates enforcement in the attorney general with penalties up to $25,000 per violation. In the absence of a federal standard, the result is a state-by-state patchwork.
The SEC cancelled its own crypto vote, then approved the proposal four days later. Here is what 'Regulation Crypto Assets' actually says
After cancelling a public meeting, the SEC approved a crypto-offering proposal on 18 August by seriatim vote. The text is now public: a tiered fundraising exemption up to $75 million, a $5 million startup path, and a safe harbour that turns on ceasing 'essential managerial efforts.' A proposal, not law, unlikely before 2027.
The SEC is about to start writing crypto's rulebook itself, with Congress stalled
On Friday the SEC votes on whether to propose 'Regulation Crypto,' a set of token-fundraising rules it has been sketching since 2025, now arriving as the CLARITY Act sits stalled in the Senate. Here is what tomorrow's vote actually decides (less than the headlines suggest), what the rules would do, and why nothing changes on Friday.
OpenAI's 'Jalapeño' chip beat Nvidia on efficiency: what it actually means
OpenAI showed off its first in-house AI chip and said it does more work per watt than Nvidia's Blackwell. The claim is real, but the fine print matters: these are OpenAI's own benchmarks, against Nvidia's older memory, from a chip that barely ships yet.
An AI designed protein binders largely on its own, and independent labs confirmed they actually stick
Anthropic says Claude ran a protein-binder campaign with little human input. Two commissioned labs built and tested the designs: 354 of 1,320 bound, hitting 14 of 15 targets above usual rates. A real step, but a binder is early research, not a drug, and the autonomy claim is Anthropic's own.
OpenAI built an AI that's more willing to hack, and locked it to defenders
OpenAI released a model trained to refuse fewer offensive-security requests, from finding zero-days to building exploit chains, and made it available only to vetted security firms. It is a genuinely careful move dressed in an alarming headline. Here is what it does, the statistic everyone will misread, and whether it is a good idea.
Meta just gave away an AI you can run yourself, and reopened the biggest fight in AI
Meta's new Muse Glimmer is a capable model, freely licensed, that runs on your own computer with no API and no internet. It is the loudest shot yet in the open-versus-closed battle. Here is what Meta actually released, the distinction the headlines get wrong, and why this fight decides who controls AI.