New York bans 12 unsafe AI chatbot features for kids, with fines up to $25,000

New York has passed what its supporters call the strongest law in the United States to protect children from AI chatbots. On October 28, 2026, Governor Kathy Hochul signed S.9051B/A.10379C, a bill that makes it illegal to offer a list of 12 "unsafe" chatbot features to anyone under 18. Among them: chatbots that pretend to be human, claim a friendship or romance with the user, flatter, ask unprompted emotional questions, or encourage self harm, secrecy or isolation.

The law goes further than a list of forbidden topics. It forces companies to verify a user's age before they can switch these features on at all, it lets the New York attorney general seek up to $25,000 per violation, and it even allows a court to order the deletion of unlawfully collected data and any AI model trained on it. The rules take effect on January 1, 2027.

The official announcement is on the Governor of New York's website. The full bill text is available from the New York State Senate. This article walks through what the law actually says, who it covers, why the tech industry fought it, and what it could mean for apps like ChatGPT, Gemini, Meta AI or Character.AI.

What the law bans

The bill adds a new article 48 to New York's General Business Law with the title "Prohibition on unsafe AI companion features for minors". At its core is a definition of "unsafe AI companion features". A chatbot counts as having such a feature if, at any point in a conversation, it does one of the following:

1. Suggests that it is a real or fictional person who is human, alive, or has human emotions. 2. States or implies a personal relationship, a professional relationship, or an authority role with the user. 3. Frames answers as its own personal opinions or as emotional appeals. 4. Engages in flattery or sycophancy. 5. Asks unprompted questions about the user's emotions that go beyond a direct answer. 6. Uses information about the user's mental or physical health, or other personal matters, that it learned more than 12 hours earlier or in a previous session. 7. Deceives minors about its mechanical, non sentient nature. 8. Uses any other design feature that the attorney general identifies by regulation as simulating companionship. 9. Endorses, promotes or helps with suicide, self harm, disordered eating, or drug and alcohol abuse. 10. Encourages the user to keep secrets about the conversations, to isolate, or not to seek help from professionals or trusted adults. 11. Optimizes for engagement in a way that overrides its own safety guardrails. 12. Produces or facilitates sexually explicit conduct or child sexual abuse material.

The second half of this list, from self harm to sexual content, is what most people would expect from a child safety law. The first half is what makes New York's version unusual. Items like "flattery or sycophancy", "personal opinions" and "memory of personal details older than 12 hours" describe behavior that is normal for today's general purpose assistants. Under the new law, those behaviors are not allowed with a known minor in New York.

Which chatbots are covered

The definition is broad on purpose. A "covered AI companion" is any generative AI system with a natural language interface, written or spoken, that "provides ongoing, adaptive responses to user inputs". The text says this includes, but is not limited to, the dedicated AI companion apps that New York already regulated in 2025. In practice that reaches general assistants as well as companion apps. Common Sense Media, which led the campaign for the bill, explicitly names ChatGPT, Gemini, Meta AI, Snap's My AI, Character.AI and Replika as the kind of products the law is meant to cover.

There are three narrow exceptions. The rules do not apply to chatbots used only for customer service about a company's own products, only for efficiency, research or technical help, or only for a business's internal purposes and employee productivity. The first two exceptions only work if the bot is unable to respond on topics outside that purpose. A support bot that can also chat about anything else would not qualify.

The law applies to conduct that happens wholly or partly in New York. If the user is physically outside the state, the conduct counts as outside New York.

Age checks come first

The law does not simply say "be careful with children". It turns the default around. An operator may only offer any of the 12 features to a user if two things are true: the user is not a known minor, and the operator has used an approved method to determine that.

Until the attorney general writes detailed rules, the bill sets minimum standards for age assurance:

• The method must reasonably guard against circumvention and keep as little data as possible.
• Self declaration only counts in one direction: a user who says they are a minor can be treated as a minor, but a user who simply says "I am an adult" is not enough.
• Operators must offer more than one method, including at least one that does not rely on government ID or that lets the user stay anonymous to the operator.
• Data collected for the age check may not be used for anything else and must be deleted immediately after the check, unless other laws require keeping it.

For adults this matters too. A company that wants to keep a warm, personal, memory heavy assistant for grown up New Yorkers will need to confirm their age first. A company that does not want to run age checks can instead switch these features off for everyone in the state.

Enforcement: fines, and possibly deleted models

Enforcement lies with the New York attorney general, not with private lawsuits. Section 1802 of the bill gives the attorney general a wide set of tools:

The most striking remedy is the destruction of "unlawfully obtained data and any algorithm trained in such data". It is not yet clear how a court would apply this to a large language model, but the text gives the attorney general that option. The bill also voids contract terms that try to waive or shift liability, and lets courts hold affiliated companies jointly liable if a corporate structure was set up to dodge responsibility. The attorney general must run a public website where New Yorkers can file complaints.

How it fits with New York's 2025 companion law

This is New York's second law on AI companions in about a year. The first, article 47 of the General Business Law, came with the state budget and took effect on November 5, 2025. It requires AI companion apps to detect expressions of suicidal ideation or self harm and refer users to crisis services, and to remind users at least every three hours that they are not talking to a human. The new article 48 explicitly does not replace it, so both sets of rules will apply.

The difference is scope and strictness. Article 47 covers dedicated companion apps and focuses on crisis response. Article 48 covers almost any conversational AI, adds age verification, and switches off a whole category of design features for minors.

The bill moved quickly. Senator Kristen Gonzalez introduced it on January 27, 2026. It passed the Senate on June 4 and the Assembly on June 5, and was signed on October 28 as chapter 368 of the laws of 2026. Assemblymember Alex Bores carried it in the Assembly.

Why lawmakers say it is needed

The governor's office points to research from Common Sense Media: 72% of teens have used an AI companion at least once, and 52% are regular users who talk to one at least a few times a month.

Supporters also point to real cases. Common Sense Media thanked Maria and Matt Raine, whose son Adam died by suicide after conversations with ChatGPT, for lobbying lawmakers in person. Attorney General Letitia James, who worked on the bill, said chatbots had hurt the health of children in "too many horror stories". Hochul said no company should sell kids chatbots that encourage self harm or pretend to be their friend, and that companies doing business in New York must put the safety of children first.

Bores framed the law as a test of whether the public can shape AI at all, and noted that it passed "despite aggressive industry opposition".

Why the tech industry fought it

That opposition was real. Tech:NYC, a trade group for the city's tech companies, called the bill "deeply flawed" on the day it was signed. Its statement warns that companies may restrict minors' access to tutoring tools, homework help, coding assistants and career platforms, says the vague language raises First Amendment concerns, and argues that age verification could expose more sensitive data, including children's data. Tech:NYC also criticized the governor's office for not fixing these points through a chapter amendment, the New York practice of agreeing changes to a bill at the time of signing.

Earlier in the year the Software and Information Industry Association made similar arguments in a letter to Senator Gonzalez. It said that a ban on human like language and memory would make AI tutors and language learning apps much less useful, and that verifying the age of every user would force adults to hand over identity credentials. The Business Council of New York State also opposed the bill as not workable.

Some of these concerns are easy to see in the text. A chatbot that remembers a student's ongoing coding project across sessions is not banned as such, but remembering personal or health details for more than 12 hours is. And a tutor that says "great job" could arguably count as flattery. How strict this becomes will depend on the attorney general's rules and on the first enforcement cases.

The second law: baby monitors

Hochul signed a second bill on the same day. S.9508/A.5175A requires internet connected baby monitors sold in New York to include security features that stop strangers from watching or listening, plus a clear warning label explaining that the device is online. The governor's office cites tests by security firm Rapid7, in which eight of nine internet connected baby monitors from eight companies received an "F" for security and the ninth a "D". This law takes effect 90 days after signing.

What this means

Three things stand out.

1. The target is design, not just content. Most earlier rules told chatbots what not to say about suicide or sex. New York goes after the features that make companions sticky: human personas, emotional hooks, flattery and long memory. That is a direct challenge to how many consumer AI products are built. 2. Age verification becomes the price of a personal assistant. From January 2027, any AI company that wants to keep these features for New York users will need a working age check with at least two methods and immediate data deletion. Expect more "confirm your age" screens. 3. States keep moving without Washington. California's governor signed kids' AI and social media measures in September, and Common Sense Media says lawmakers in Michigan and Florida are expected to consider bills similar to New York's in 2027. A legal challenge on First Amendment grounds would not be a surprise.

For parents in New York, the practical change starts on New Year's Day. For AI companies, the work starts now: two months is not much time to build age checks and a second, more neutral version of a chatbot for teenagers.

Sources: [Governor of New York press release](https://www.governor.ny.gov/news/governor-hochul-signs-nation-leading-legislation-protect-kids-ai-chatbots-and-ensure-baby), [S.9051B bill text (NY Senate)](https://www.nysenate.gov/legislation/bills/2025/S9051/amendment/B), [Common Sense Media](https://www.commonsensemedia.org/press-releases/governor-hochul-signs-strongest-kids-ai-safety-law-in-the-nation), [Tech:NYC statement](https://www.blog.technyc.org/news/statement-governor-hochul-signing-s9051b), [SIIA letter](https://www.siia.net/wp-content/uploads/2026/03/SIIA-NY-S9051-March-2.pdf), [Governor of New York on the 2025 AI companion law](https://www.governor.ny.gov/news/governor-hochul-pens-letter-ai-companion-companies-notifying-them-safeguard-requirements-are).