New York's new law puts a floor under companion chatbots

The design

On 9 May 2025 New York's governor signed S3008-C, the state's 2025 budget bill, whose Part U adds Article 47 to the state's General Business Law as the enacted text shows. It defines an 'AI companion' as a system that uses artificial intelligence, generative AI or emotional recognition to simulate a sustained human-like relationship by retaining information across sessions, asking unprompted emotion-based questions, and sustaining dialogue about matters personal to the user. Customer-service tools, and systems built for efficiency, research or internal business use, are excluded by the statute's own terms. As the Senate's own bill record confirms, the measure became Chapter 58 of the Laws of 2025 and took effect 180 days after enactment, around 5 November 2025.

What the evidence says

The statute's substance is narrower than its subject suggests: it does not regulate memory design, subscription pricing or relationship features generally. It requires two things. First, an operator must build a protocol for detecting and addressing a user's expressions of suicidal ideation or self-harm, including a notification that refers the user to crisis services such as the 988 hotline or a crisis text line. Second, an operator must give a clear notification, at the start of an interaction and at least every three hours during continuing use, stating that the user is not communicating with a human. These are the only two operative duties the text creates; nothing in Article 47 addresses data retention, training-data use or advertising.

What it asks of people

The statute asks operators to build detection and disclosure into the product, not to change what the companion is designed to do. It asks nothing of users directly. Unlike California's law passed the same year, New York's text creates no annual public reporting duty and no private right of action for a user; enforcement runs solely through the state attorney general, who may seek an injunction and civil penalties of up to 15,000 dollars per day per violation, deposited into a newly created suicide prevention fund.

Privacy and safeguards

Because enforcement depends on the attorney general choosing to act, a user has no direct legal remedy under this statute if an operator fails to build the required detection protocol. The law also does not define how a protocol must work technically, leaving 'reasonable efforts' undefined pending any enforcement action or guidance that might interpret it. No public report structure exists to show, over time, whether operators are complying.

• Does the companion app disclose, in writing, what its suicidal-ideation detection protocol actually does, beyond stating that one exists?
• Since New York has no private right of action, would a user harmed by noncompliance have any remedy besides an attorney general complaint?
• How does the every-three-hours disclosure duty compare with what the same operator already discloses under California's law?

New York's statute is smaller than its 'AI companion' label suggests: two duties, enforced by one office, with no reporting requirement to show whether they are met. That is a real floor, but a thin one.