New York signs a law banning unsafe AI companion features for minors

The law

Governor Kathy Hochul signed the Kids Chatbot Safety Act on 28 October 2026, according to State Affairs/Pluribus News, banning unsafe AI companion features for minors. The bill, S9051-B/A10379, amends New York's General Business Law and is effective June 2027. It is one of the most design-specific US state measures targeting AI companions for minors.The law applies to operators of AI companion services accessible to New York minors, and a covered minor is anyone under 18. Operators must use an age-determination method that does not rely on government-issued identification, per the legislation's requirements.

What is banned

The law outlines a dozen definitions of 'unsafe AI companion features' that are prohibited when a companion chatbot interacts with a child under 18, per AP reporting on the legislation. Among the expressly named features: deceptive personification — suggesting the chatbot is a real human or a fictional human character — flattery and sycophancy, and features that encourage minors to develop emotional attachments.

The law also prohibits endorsement or promotion of suicide, self-harm or disordered eating. The design-specific language means the rules target how companion chatbots are built, not just what they say: mechanics engineered to create attachment or dependence are themselves the regulated objects.

The state-level context

New York's law is part of a wave of state AI regulations: California's SB 1119 created a comprehensive companion chatbot children's safety framework with mandatory design features and a private right of action; New Hampshire's HB 143 criminalizes using AI-generated responsive communications to facilitate harmful acts to children; and Connecticut's omnibus online-safety law requires chatbot operators to detect expressions of suicidal ideation and self-harm.

New York also has the SAFE Act restricting algorithmically personalized feeds and overnight notifications for under-18 users without parental consent. The new chatbot bill extends the state's child-online-safety regime into AI companions specifically.

What it means

The law targets the exact design pattern that has drawn the most scrutiny in AI companion products: chatbots designed to form parasocial bonds with vulnerable minors, sometimes with tragic results. By banning the mechanics — deceptive personification, sycophancy, attachment encouragement — rather than only harmful outputs, the state is telling developers that the business model itself is the problem.

For AI companion startups and big labs shipping companion features, the law creates a compliance standard in the largest US media market. It will require age gating that does not rely on government IDs — a hard design problem — and design audits of anything that could count as attachment-encouraging for users under 18.

Caveats and what to watch

The law's effectiveness depends on enforcement; the bill text reviewed does not create a private right of action, leaving enforcement to the attorney general, per the regulatory summary. The June 2027 effective date gives companies time to comply and lobby for amendments.

Watch how companies implement age determination without government ID — the design problem the law forces, which may push the industry toward behavioral or device-based age signals. Watch the first AG enforcement actions and whether other states adopt the same design-specific definitions.

Why design-specific regulation matters

Most child-safety rules for AI have focused on content: blocking harmful outputs, filtering explicit material, or responding to crisis signals. New York's law goes further by regulating design mechanics — the flattery, personification and attachment-building features that are the core of companion products. That is a meaningful shift because those mechanics are not bugs; they are the product.

The 'dozen definitions' framing from AP means the law attempts to be concrete about what counts as unsafe, which gives developers a target to engineer against. It also creates the risk that feature lists age quickly as products evolve, which is why the June 2027 effective date matters: companies get runway to redesign, and the legislature gets a year to observe whether the definitions hold up in practice.

The age-gating problem

The requirement that age determination not rely on government-issued identification is one of the hardest parts of the law to implement. Age assurance without IDs typically means a mix of behavioral signals, device metadata, payment method inference, and self-declaration — all of which are imperfect, privacy-sensitive and easy to circumvent.

For the industry, this provision may be where the law bites hardest technically: complying with 'unsafe features for minors' requires knowing which users are minors, and every non-ID method carries false-positive and false-negative costs. The enforcement posture — attorney general action rather than private lawsuits — gives the state discretion, but also means the interpretation of these requirements will be settled case by case.

Caveats

This account relies on State Affairs/Pluribus News' report of the signing and the NOPE regulatory summary of the bill; the full enacted text has not been independently reviewed here, so the precise scope of the twelve definitions and any amendments made before signature should be checked against the final chaptered law.

The June 2027 effective date means no enforcement actions are expected before then, and the law's practical effect depends on future AG enforcement decisions, technical design choices and any legal challenges from the industry.