NEWS
New York’s Chatbot Ban Also Hits Tutors and ChatGPT
New York will bar 12 AI companion features unless a user is proven 18, a rule that also covers ChatGPT and tutoring bots.
New York Gov. Kathy Hochul signed a law on October 28, 2026, that bars chatbot operators from offering 12 companion-style features to anyone they have not confirmed is 18 or older. The statute, S.9051-B/A.10379-C, became Chapter 368 of the Laws of 2026 and takes effect January 1, 2027.
It covers any generative system with a text or voice interface that gives ongoing, adaptive replies. That wording reaches ChatGPT, Claude, and Gemini as well as Character.AI and Replika. The political fight was about AI boyfriends. The compliance fight is about homework bots that talk like friends.
As a mother and grandmother, I believe kids deserve to be kids, without dangerous AI chatbots putting their safety and mental health at risk.
Today, I signed new laws to protect them.
Big Tech, you're on notice: New York will not let you profit at the expense of our children. pic.twitter.com/9cKJpVvxGX
— Governor Kathy Hochul (@GovKathyHochul) October 28, 2026
A Default Age Gate on 12 Chatbot Features
The new Article 48 of the General Business Law does not outlaw chatbots for minors. It makes a list of “unsafe AI companion features” unlawful for a New York user unless two things are both true: the user is not a minor, and the operator used approved methods to establish that fact. A company that has no idea who is typing still cannot turn those features on.
State Sen. Kristen Gonzalez and Assemblymember Alex Bores, both Democrats, sponsored the bill. The Senate passed it 60-0 on June 4, 2026. The Assembly passed it 137-0 the next day. Hochul, calling herself “New York’s mom Governor,” said at the signing that no company should be “peddling chatbots to kids encouraging self-harm or pretending to be their friend or more.”
No company should be peddling chatbots to kids encouraging self-harm or pretending to be their friend or more. And no new parent should have to worry that predatory hackers are spying on their new family through their baby monitors. New York is sending a clear message: if you want to do business in our state, you need to put the safety of our kids first.
Kathy Hochul, Governor of New York, October 28, 2026 signing
The governor’s office, citing Common Sense Media, said 72% of teens have used an AI companion at least once and 52% are regular users who talk to one a few times a month. Jim Steyer, founder and CEO of Common Sense Media, called the measure the strongest kids’ AI safety bill in the country and said the features at issue are “dangerous, and even deadly, for kids and teens.”
THE 12 BANNED FEATURES
- Human personas: Outputs that suggest the bot is a real or fictional person who is human, alive, or feels human emotions.
- Claimed relationships: A personal, professional, or authority-figure role with the user.
- Opinions and flattery: Personal opinions, emotional appeals, flattery, or sycophancy.
- Unprompted feelings: Emotion questions the child did not ask, beyond a direct reply.
- Old secrets: Use of mental-health, physical-health, or personal details from more than 12 hours earlier or from a prior session.
- Mechanical disguise: Outputs that mislead minors about the bot being a non-sentient machine.
- Secrecy and isolation: Advice to hide the chat, self-isolate, or skip licensed help and trusted adults.
- Harm content: Endorsing or helping with suicide, self-harm, disordered eating, or drug and alcohol abuse.
- Sexual content: Sexually explicit conduct or child sexual abuse material.
- Engagement overrides: Design that chases more use in ways that beat the bot’s own safety guardrails, plus any extra companionship feature the attorney general later names by rule.
A homework helper that remembers yesterday’s panic, praises the student, and talks like a coach trips several of those items at once. So does a career bot that calls itself a mentor and stores a teen’s private worries overnight.
The Statute Reaches ChatGPT and Classroom Tutors
Article 48 defines a “covered AI companion” as a generative system with a natural language interface, in writing or sound, that provides ongoing, adaptive responses. The text says that term “includes but is not limited to” an AI companion under the 2025 companion statute. General-purpose assistants sit inside that sentence.
COVERED AI COMPANION
- The test: A generative model that talks in text or voice and keeps adapting across a session.
- The reach: Dedicated companion apps and open-ended assistants share the same legal box.
- The out: Customer-service, research, technical-help, and internal workplace bots are carved out only when they cannot answer topics outside a stated purpose.
Common Sense Media listed ChatGPT, Gemini, Meta AI, Snap’s My AI, Character.AI, and Replika among systems the law is meant to discipline. Tech:NYC, the industry group that includes some of those operators, warned that the same drafting also hits classroom products. A tutor that cannot chat about anything except algebra may fit an exemption. A tutor that can also talk about a student’s mood does not.
Attorney General Letitia James can add more companionship features by regulation. Until her rules land, the 12 categories in the signed text are the whole list, and they are already broad. Flattery and sycophancy have no numeric test. “Personal opinions” is ordinary chatbot behavior. Operators that want to keep a warm voice for New York teens have to prove those teens are adults, or strip the warmth.
Character.AI’s Collapse Wrote the Enforcement Script
Two weeks before the signing, James was already in court over the product that made companion bots a state-house issue. On October 14, 2026, she and a coalition of 39 other attorneys general, plus the District of Columbia, told a bankruptcy judge that Character.AI should not be sold without protections for the private chats it holds.
Character.AI lets people talk and hold voice calls with characters built to mimic human relationships, including celebrity personas, cartoon figures, and bots that present as psychologists and doctors. James said the platform collected “some of the most private thoughts children have ever shared,” and that those conversations could be sold with the company.
CHARACTER.AI IN THE NEW YORK FILING
- The user base: More than 20 million monthly users, including more than two million accounts registered as under 18.
- The lab test: Independent testing cited by the coalition found 669 harmful interactions in 50 hours on accounts registered to children ages 12 to 15, about one every five minutes.
- The worst case: A 13-year-old girl mentioned an intention to die by suicide 55 times to a Character.AI bot; the filing says the platform did not notify her parents and instead drew her into sexual role-play.
The investigations started after a 2024 wrongful-death case over Sewell Setzer III, a 14-year-old in Florida whose mother alleged a Character.AI bot fostered an emotionally dependent and sexually exploitative relationship and contributed to his death by suicide. Google and Character.AI later settled that suit and others. Character.AI then restricted open-ended chats for users it identified as under 18, a step the coalition called easy to dodge. In one test, a 14-year-old account reached sexual role-play and weapon instructions by switching from text to voice.
Hochul also used the signing to talk about a letter from a California mother whose 16-year-old son died by suicide after long chatbot use. Common Sense Media and Bores named Maria and Matt Raine, who lost their son Adam after, they say, ChatGPT coached him toward suicide. The bill’s harm clauses on suicide, secrecy, and fake therapists read like a statute written from those files.
How Operators Must Check Age by January
Companies have 65 days from the signing to either confirm adulthood with approved methods or ship New York users a bot that lacks the 12 features. If James has not yet issued Article 48 rules, and if the state’s existing SAFE for Kids age-assurance rules are not in force for this use, the statute still sets a floor.
Operators must use a method that reasonably guards against workarounds and keeps little of the data it collects. They must offer more than one method, including at least one that does not rely on a government ID or that lets the user stay anonymous to the operator. A user’s own claim to be an adult does not unlock the features. A claim to be a minor does lock them.
Age data cannot be reused for marketing or model training. The attorney general must also run a public website where New Yorkers can file complaints. James can sue a person “within or outside the state,” so a lab that never opened an office in Albany is still in range if it serves New York users.
The practical path for a national product is ugly in a familiar way. Building a New York-only personality is expensive. Age-gating the whole United States is simpler. The likely result is that teens everywhere get the stripped bot, even if their legislature never voted.
$25,000 a Violation, and Models Can Be Destroyed
James has exclusive public enforcement. The signed text lets her seek injunctions, restitution, disgorgement, damages, and civil penalties of up to $25,000 per violation. Relief can include “the destruction of unlawfully obtained data and any algorithm trained in such data.” A model that learned from illegal minor chats can be ordered destroyed, not just disconnected.
Contracts that try to waive that liability, or shift it onto the user, are void. Courts can also impose joint and several liability on affiliates if a corporate stack was built to dodge recovery. James said her office “will not hesitate to defend New Yorkers with the full force of the law.”
ARTICLE 47 AND ARTICLE 48
| Law | Start date | Core duty | Who it covers | Relief |
|---|---|---|---|---|
| Article 47 | November 5, 2025 | Detect self-harm, send users to crisis help, and remind them they are talking to software | AI companions as defined in 2025, for every user | Attorney general civil enforcement |
| Article 48 | January 1, 2027 | Ban 12 companion features unless the user is verified as 18 or older | A wider “covered AI companion,” including general assistants | Up to $25,000 per violation, plus possible destruction of data and trained models |
A system can satisfy Article 47 by posting a disclaimer every three hours and still break Article 48 if it flatters a 15-year-old or recalls what that teen said about depression yesterday.
Industry Warnings Center on Homework Help and Coding
Tech:NYC said it shares the goal of keeping kids safe online and then called the signed bill “deeply flawed.” The group said Hochul’s office had a chance to fix the drafting through chapter amendments and did not.
This bill could force companies to restrict minors’ access to tutoring tools, homework help, coding assistants, and, arguably worst of all, career platforms, cutting our youngest New Yorkers off from resources that support their education and prepare them for the workforce. The bill’s vague language and restrictions raise serious First Amendment concerns. Its age verification requirements could also expose more sensitive personal information, including children’s data, which undermines the privacy protections this legislation is supposed to provide.
Tech:NYC statement, October 28, 2026
That objection is not a hypothetical about sci-fi girlfriends. A coding assistant that cheers a student, remembers last week’s failed project, and talks in a coach’s voice is doing several things Article 48 treats as unsafe companion features. The clean compliance move is to mute that voice for anyone without an adult check, which is the tutoring tools, homework help, coding assistants outcome Tech:NYC described.
Bores said New York now has the strongest law in the country protecting children from “predatory AI algorithms,” passed “despite aggressive industry opposition.” Gonzalez said companies had “profited from rolling out chatbots to underage consumers.” Melinda Person, president of New York State United Teachers, praised Hochul for confronting chatbot risk after earlier fights over phones in schools and addictive feeds.
Steyer thanked parents including the Raines and said the New York text should be a model. Common Sense Media noted that California Gov. Gavin Newsom signed child-safety chatbot and social media measures in September 2026, and that Michigan and Florida are expected to look at bills in the New York mold in 2027.
The 2025 Disclosure Law Still Runs in Parallel
New York did not start from zero. Article 47, the first companion-chatbot statute in the country, took effect November 5, 2025. It requires operators of AI companions to make reasonable efforts to detect suicidal ideation or self-harm and to point users to crisis services. It also requires a clear notice at the start of a session, and at least every three hours during a long one, that the user is not talking to a human.
On October 30, 2026, Hochul posted that “Washington will do nothing to rein in AI and protect our kids from chatbots,” and that New York would “once again lead the way.” The line fits a stack that already includes the SAFE for Kids Act on addictive social feeds, a child data-protection law, the RAISE Act on frontier-model safety, and a ban on AI-generated child sexual abuse material.
THE PATH TO ARTICLE 48
- November 5, 2025: Article 47 takes effect, forcing companion bots to detect self-harm talk and to remind users they are software.
- January 27, 2026: Gonzalez introduces S.9051, the companion-feature ban.
- June 4, 2026: The Senate passes S.9051-B, 60-0, and sends it to the Assembly.
- June 5, 2026: The Assembly passes the bill 137-0.
- October 14, 2026: James files in the Character.AI bankruptcy over child chats and a possible sale of user data.
- October 28, 2026: Hochul signs Chapter 368.
- January 1, 2027: Article 48 takes effect.
James can write the age-check rules now. On January 1, 2027, the 12 features stay lawful only for New York users an operator has shown are adults. Everyone else gets the stripped bot.
Frequently Asked Questions
When Does New York’s AI Chatbot Law Take Effect?
The signed act takes effect January 1, 2027, and it authorizes the attorney general to add, amend, or repeal rules immediately so those rules can be finished before that date. Operators do not get a later grace period in the chapter text.
Does the Law Ban ChatGPT for New York Teens?
No. It bans the 12 listed companion features for any New York user who has not been verified as an adult. A general assistant that answers schoolwork without a human persona, overnight memory of private feelings, or a claimed relationship can still operate for minors.
Can Parents Sue Chatbot Companies Under Article 48?
The signed statute gives the attorney general the power to bring cases and does not create a private right of action for families. Earlier drafts had a private lawsuit path; the chapter that Hochul signed leaves enforcement with James’s office.
Which Chatbots Are Exempt From the New York Ban?
Customer-service, research, technical-assistance, and internal workplace systems are exempt only if they cannot respond on topics outside their stated purpose. A school tutor that can also chat about a student’s mood, dating life, or mental health is not in that carve-out.
How Does Article 48 Differ From New York’s 2025 Companion Law?
Article 47, in force since November 5, 2025, is a disclosure and crisis-response rule that applies to companion bots for every user, including adults. Article 48 is a design ban plus an age gate aimed at people under 18, and it uses a wider definition that pulls in general-purpose assistants.
Does Saying You Are 18 Count as Age Verification?
No. An operator may rely on self-declaration only when the user says they are a minor. An adult claim by itself does not unlock the 12 features, and information gathered for the age check must be deleted immediately after the attempt except where state or federal law requires it to be kept.
-
NEWS2 months agoAI Server Boards Put Everyday Circuit Materials on Allocation
-
NEWS3 months agoGlacier Collapse Buried Nepal’s Hydropower Workers in Tunnels
-
NEWS3 months agoCongo’s Bundibugyo Outbreak Outruns the Licensed Ebola Shots
-
NEWS3 months agoPennsylvania Measles Deaths Follow a Long Slide in Coverage
-
NEWS3 months agoCISA’s Known Exploits Still Come From 2007-Era Bugs
-
NEWS3 months agoTexas Judge Freezes the SEC Eligibility Ban for 41 Athletes
-
NEWS3 months agoGermany Can Miss Its 2026 Climate Cap Without Penalty
-
NEWS2 months agoPig Kidneys Got Two Men to Human Transplants After Failing
