News · 2026-10-29

New York restricts specified AI companion features for minors from January

New York Governor Kathy Hochul signed a law on October 28 restricting specified AI companion features for minors, with the new rules taking effect January 1, 2027. The measure regulates enumerated behaviors and age assurance rather than relying only on a disclosure that a chatbot is artificial. It creates a near-term product-design deadline for operators within its scope.

Key facts

• The AI measure is S.9051B/A.10379C and adds Article 48 to New York’s General Business Law.
• The effective date is January 1, 2027.
• The Attorney General may seek civil penalties of up to $25,000 per violation.
• Primary source: the enrolled S.9051B text.

The Governor’s official release and transcript describe the signing of two separate measures. One concerns AI companion behavior. The other concerns security and warnings for internet-connected baby monitors. They share a child-safety framing, but they have different requirements and effective periods. Combining them into a single generic technology law would hide what developers and manufacturers actually need to examine.

For the AI measure, the enrolled language is the controlling source. Its definition of a covered companion uses a natural-language interface and ongoing, adaptive responses and “includes but is not limited to” an AI companion. That wording makes the scope question broader than whether an application markets itself as a romantic companion. Branding alone does not settle whether a system falls within the definition or an exception.

The prohibited features include outputs suggesting that a system is human, alive, or has human emotions; specified claims of personal, professional, or authority relationships; emotional appeals, flattery, and sycophancy; and unsolicited emotion-based questioning. The statute also addresses encouragement of self-harm and other listed harmful conduct, secrecy or isolation, and engagement optimization that overrides safety guardrails. The exact enumerated wording matters more than a summary label such as manipulative chatbot.

One provision directly intersects with a central selling point of persistent agents: memory. The text includes restrictions involving use of health, well-being, or personal information obtained more than 12 hours earlier or in a prior session. This is a specific statutory condition, not a general statement that every chatbot must forget everything after 12 hours. Product teams should read that clause together with its defined scope and the broader list of restricted features.

A useful analogy is a rulebook for what a service may do with a young customer, rather than merely a label on the door. Telling the customer that the service is automated addresses identity. Restricting relationship claims, emotional pressure, and certain uses of remembered information addresses behavior. Our lessons on sycophancy and agent memory explain why those are technically distinct design questions.

Where the prohibition applies, operators must use permitted methods to establish that a user is not a covered minor. Pending applicable regulations, they must offer more than one age-assurance route, including one that does not depend on government identification or allows anonymity from the operator. Information collected for determining age must be used for that purpose and deleted immediately after an attempt, subject to the statute’s legal-compliance exception.

Those provisions create a real implementation tension: a system needs enough assurance about age while limiting collection and retention of identifying information. The legislation states the requirements; the dossier does not establish that every operator already has a tested way to meet them. It also does not establish a broad expert consensus about the law’s effects. The governor’s child-safety rationale is an institutional position, not an independent impact evaluation.

The statute provides exceptions for specified constrained customer-service, efficiency, research, technical-assistance, and internal productivity systems. It also expressly leaves the earlier Article 47 intact. That makes it inaccurate to describe the new measure simply as the older companion disclosure and crisis-response rules being signed again.

The distinction between a defined feature restriction and a ban on all useful conversational systems is important. The enrolled text contains both broad definitions and specific exceptions. Neither the most expansive interpretation nor the narrowest product-marketing description can replace reading those provisions together.

The separate baby-monitor law requires security features and written warnings for new covered devices broadcasting through an internet connection and takes effect on the 90th day after becoming law. For AI operators, the immediate story remains Article 48: a dated behavioral and age-assurance obligation. The caveat is that a plain-language news account cannot resolve every coverage or implementation question in the enrolled text.

Primary source, verified: read the paper →

Ground Truth. (2026, October 29). New York restricts specified AI companion features for minors from January. Ground Truth. https://groundtruth.day/news/new-york-ai-companion-minor-safety-law.html

@misc{groundtruth:new-york-ai-companion-minor-safety-law,
  title  = {New York restricts specified AI companion features for minors from January},
  author = {{Ground Truth}},
  year   = {2026},
  month  = {oct},
  url    = {https://groundtruth.day/news/new-york-ai-companion-minor-safety-law.html}
}

Topics: policy · child-safety · ai-companions · new-york · privacy