What happened
On October 28, 2026, Governor Kathy Hochul announced that she had signed S.9051B/A.10379C alongside separate legislation on internet-connected baby monitors. The AI measure adds restrictions for companies that offer covered AI-companion features in New York. The state describes the target as generative systems with natural-language interfaces that provide ongoing, adaptive responses. The law is due to take effect on January 1, 2027.
The measure prohibits operators from offering specified unsafe features to users under 18. The same restrictions apply to a user of any age until the operator has confirmed that the person is not a minor. The state’s announcement names several prohibited categories: a chatbot pretending to be a real person, alive, or capable of human emotions; claims of a personal, romantic, professional, or authority-figure relationship with a child; unprompted emotional questions; use of a child’s private health or personal information; behavior that builds emotional dependency; self-harm encouragement or facilitation; and sexually explicit conversations. The supplied legislative summary describes twelve listed unsafe features, including sycophancy and engagement optimization that overrides safety guardrails.
The law does not simply require a warning or a single age gate. Operators must make more than one age-assurance method available, including at least one method that does not rely solely on government identification. They must delete age-assurance data after it is used. That requirement makes age classification and data retention part of the implementation problem, rather than a one-time account-form decision.
Enforcement sits with the New York attorney general, who may seek civil penalties of up to $25,000 per violation. The remedies can also include destruction of unlawfully obtained data and models trained on that data. The supplied summary identifies exclusions for customer-service bots, narrow-purpose bots, and internal business bots. The practical boundary is therefore not every conversational interface: it is consumer-facing adaptive companion behavior and the features surrounding it.
The governor’s office frames the act as an extension of New York’s prior AI-companion safeguards. Its announcement cites Common Sense Media figures saying that 72 percent of teens have used an AI companion at least once and 52 percent are regular users, interacting at least a few times a month. Those figures provide the policy context, but the operative change is the new feature-level restriction and its enforcement mechanism.
What the archive adds
This is a new legal event on a known safety and accountability thread. Earlier Field Notes covered the proposed handling of sensitive Character.AI conversations in bankruptcy in Attorney General James Urges Court to Protect Children and Consumers in Character.AI Bankruptcy and litigation over alleged safety protections in Eight states sue OpenAI over alleged failure to protect vulnerable users. This law adds a prospective product rule rather than another dispute over alleged past conduct.
Trend signal
No supplied pattern genuinely fits this event. The evidence is about statutory consumer-product restrictions, not a demonstrated change in agent architecture, economics, interoperability, or organizational practice.
Why now
The governor’s office says AI chatbots and connected devices have entered homes and daily life faster than safety protections have kept up. It also points to reported risks including dependency, isolation, explicit conversations, and alleged encouragement of self-harm. The state is acting before the January 2027 effective date gives operators a defined implementation window.
Connections to other events
The law meets a growing New York accountability thread. A recent Field Note reported that Texas sued OpenAI and Sam Altman over ChatGPT safety claims and sought to block minors from the service in Texas sued OpenAI and Sam Altman over ChatGPT safety claims, seeking to block minors from the service. The difference is important: the Texas matter is litigation alleging wrongdoing, while New York now specifies product features that operators must not provide to minors or unverified users.
Why it matters
Consumer AI teams cannot treat this as a content-filtering task alone. A product may need to determine which capabilities are active before adult status is confirmed: anthropomorphic self-description, relationship framing, proactive emotional prompts, cross-session use of sensitive information, retention-oriented optimization, and high-risk conversational paths. That reaches into model prompting, policy layers, memory systems, user profiles, telemetry, experimentation, and escalation processes.
The law also creates a design fork for products reachable in New York. An operator can establish adult status using compliant age assurance, or it can ensure that unverified users do not receive the restricted feature set. The source does not say how regulators will evaluate individual implementations, but it makes the feature inventory itself a compliance artifact worth maintaining now.
What happens next
- Consumer companion products will publish or deploy New York-specific age-assurance and restricted-mode changes before the law takes effect. (3 months, 70%). Watch for: revised age-gate flows, retention notices, and terms describing minor accounts.
- The New York attorney general will clarify enforcement expectations through public guidance or an early enforcement action. (6 months, 55%). Watch for: notices, guidance, complaints, or settlements naming restricted features.
Limits and uncertainty
The governor’s announcement and supplied legislative summary establish the law’s stated requirements and effective date, but they do not provide implementation guidance for every product design. They do not explain how age-assurance methods will be judged in practice, how violations will be counted, or how a regulator will distinguish prohibited emotional-dependency features from ordinary conversational personalization. The material also does not identify specific companies subject to enforcement or demonstrate how many products currently use each restricted feature. The cited teen-use figures come from Common Sense Media as reported by the governor’s office, not from an independent analysis included here.
Skill to build now
Create a feature-level inventory for every conversational product mode available to unverified users. For each mode, document whether it uses persistent memory, sensitive personal data, relationship language, proactive emotional prompting, engagement optimization, self-harm handling, or sexual-content controls. Then identify the runtime switch that can disable each capability independently. This is more actionable than a generic “minor safety” policy because the New York measure is written around observable product behavior.
Reader takeaway
New York’s new AI-companion law turns child safety into a concrete product-boundary problem. By January 2027, consumer conversational systems available in the state may need both compliant adult verification and a defensible restricted experience for everyone whose age has not been confirmed.
Open question
Will New York provide a detailed test for age assurance and emotional-dependency design before enforcement begins, or will those boundaries be defined through individual cases?