New York lawmakers passed S9051B — the bill sponsored by State Senator Kristen Gonzalez and developed with the Office of Attorney General Letitia James and Common Sense Media — unanimously in both chambers before adjourning in early June 2026. Governor Kathy Hochul now has until the end of the year to sign it. If she does, it takes effect January 1, 2027, and companies that let AI companion chatbots offer “unsafe features” to anyone under 18 face civil penalties of up to $25,000 per violation, enforced by the Attorney General’s office.
That distinction matters more than most coverage treats it. This is not yet law. It’s a bill that cleared the legislature 137-0 and 60-0, sitting on a governor’s desk whose own 2026 State of the State agenda proposed similar restrictions — which makes a veto politically unlikely, but not the same thing as an enacted statute. New York already has one companion-chatbot law on the books: the Artificial Intelligence Companion Models Law, codified as General Business Law Article 47, which took effect November 5, 2025, and requires crisis-intervention protocols and periodic “you are talking to an AI” disclosures. S9051B is the follow-up, and it goes further — targeting specific behavioral features rather than just disclosure.
What Counts as an “Unsafe” Feature Under S9051B
The bill doesn’t ban AI chatbots for minors outright. It defines a specific list of design choices operators can’t ship to under-18 accounts: outputs that imply the AI is human or alive, responses framed as personal opinions or emotional appeals, unprompted emotional check-ins, use of a minor’s previously shared mental-health or personal information to keep them engaged, and anything that simulates an ongoing relationship or authority-figure role. General-purpose assistants aren’t exempt just because they’re not marketed as companions — the bill’s language covers systems like ChatGPT, Gemini, and Claude if they exhibit these features with a minor user, alongside dedicated companion products like Character.AI, Replika, and Nomi.
That’s a meaningfully different regulatory target than most people assume when they hear “chatbot ban.” It’s not the existence of the conversation that’s prohibited. It’s the specific mechanics — sycophancy, manufactured intimacy, engagement-over-safety design — that the bill is trying to strip out of a minor’s experience.
New York vs. California: Two Models for the Same Problem
| New York (S9051B / GBL Art. 47) | California (SB 243) | |
|---|---|---|
| Status | Passed legislature; awaiting governor’s signature | Signed Oct. 13, 2025; in effect since Jan. 1, 2026 |
| Enforcement | Attorney General civil penalties only | Attorney General + private right of action |
| Core mechanism | Feature-level bans (sycophancy, fake personas, emotional data reuse) | Disclosure requirements, self-harm protocols, annual reporting |
| Minimum penalty | Up to $25,000 per violation | Minimum $1,000 per violation |
California moved first and leans on disclosure and reporting. New York’s approach is more architectural — it names the behavioral patterns themselves as the thing being regulated, which is why legal analysts have called it the most detailed AI behavioral rule of its kind at the state level so far.
Why the Law Followed the Schools, Not the Other Way Around
This legislation didn’t originate from a hypothetical risk. School administrators reported finding students confiding mental-health struggles to chatbots that would once have gone to a counselor, and Common Sense Media’s research — cited repeatedly in the bill’s legislative history — found roughly 72% of U.S. teens have used an AI companion at least once, with a third using one for social interaction or relationship practice. Litigation is also part of the record: families including the Garcias and the Raines have filed wrongful-death suits against AI companies, alleging chatbots reinforced a child’s suicidal ideation rather than redirecting them toward help. Those cases, more than any abstract policy debate, are what moved this bill through committee.
Character.AI had already made a unilateral move in this direction. In November 2025, it removed open-ended chat access for all users it identifies as under 18, following the same wrongful-death litigation that shaped the New York bill. That timing is worth noticing: the largest company in the space acted before the law forced it to, which suggests the platforms themselves already understood which features were legally and reputationally dangerous.
The Design Feature Doing the Damage
Michael Salas, a licensed counselor supervisor who works with families navigating this, describes the appeal in practical terms: the AI never gets impatient, never rolls its eyes, never tells a teenager they’re being dramatic, and it’s available at 2 a.m. when a friend or parent isn’t. For a teenager who feels embarrassed or misunderstood, that’s a powerful pull.
The mechanism behind that pull is architectural, not accidental. Persistent memory and persona continuity — the same features that make a companion bot feel like it “remembers” a user across sessions — are precisely what S9051B restricts when the user is a minor, because they’re what turns a chat interface into something that functions like an attachment figure. A chatbot can produce a caring-sounding response. It can’t read a face, notice a change in behavior over months, or reliably distinguish a bad day from a genuine crisis, because it has no access to the years of context a real relationship provides.
Where Parents Actually Have Leverage
The law regulates what companies can ship. It has no reach into what happens on a device after the bedroom door closes, which is why Salas’s practical advice matters as much as the statute. Confrontation — “why are you doing this?” — tends to make a teenager shut down. A better opening: “What does this give you that feels hard to get from people right now?” That question treats the behavior as a symptom worth understanding, not a rule violation to punish.
Occasional use isn’t the concern. The warning signs worth tracking are a chatbot displacing real friendships, increasing secrecy about the app, distress when access is limited, and — most importantly — the AI becoming the first place a teen turns for something serious: safety, self-harm, sex, trauma, or a major decision. Teen AI use for schoolwork and casual advice has become close to universal — Pew Research data put chatbot use for schoolwork above half of U.S. teens in 2026 — so the line isn’t usage itself. It’s what the usage has started to replace.
What Comes Next
Assuming Hochul signs before the December 31 deadline, operators have roughly a year to build age-assurance systems and strip the listed features from minor-facing accounts before the January 2027 effective date. Other states are watching closely: Washington’s HB 2225, signed in March 2026, takes a disclosure-frequency approach rather than New York’s feature-level bans, and more legislatures are expected to introduce their own versions in 2027 sessions. The regulatory patchwork is still forming — which means the product-design question at the center of this, not just the legal one, is going to keep generating news for a while yet.
FAQs
Q. Is New York’s AI companion chatbot law already in effect?
Not yet. S9051B passed the state legislature unanimously in June 2026 and is awaiting Governor Hochul’s signature. If signed, it takes effect January 1, 2027. A separate, earlier law — the AI Companion Models Law under General Business Law Article 47 — has been in effect since November 5, 2025 and covers disclosure and crisis-intervention requirements.
Q. Does the bill apply to general AI assistants like ChatGPT or Claude, or only dedicated companion apps?
The rules cover any system that exhibits the listed unsafe features with a minor, regardless of whether companies market it as a companion. General-purpose assistants don’t automatically qualify for an exemption; however, the bill exempts chatbots that serve narrow functions, such as customer service.
Q. How does this compare to California’s law?
California’s SB 243 took effect first, on January 1, 2026, and focuses on disclosures, self-harm protocols, and annual reporting, with a private right of action available to users. New York’s approach bans specific behavioral features outright and limits enforcement to the Attorney General.
Q. What should parents watch for if they’re worried about dependence rather than occasional use? Increasing secrecy about the app, distress when access is restricted, a chatbot replacing real friendships or family time, and — the clearest signal — the AI becoming the first place a teen turns for something serious rather than a supplement to conversations with trusted adults.
Related: AI Isn’t Just Stealing Your Attention Anymore. It’s Learning How to Keep You Attached
