New York and California Take Diverging Paths in Second Wave of Chatbot Laws

A second generation of chatbot laws

California and New York were the first two states to enact AI companion chatbot laws in 2025. They have now become the first to adopt a second generation of legislation, only a year later, FPF notes. New York's signing brings the number of enacted state chatbot-specific laws nationwide to 19.

FPF says both states have moved beyond relatively narrow disclosure and crisis response requirements to address a wider range of harms. Both laws require age assurance and focus on minors, but the approaches differ. New York is described as restrictive by design, while California's SB 1119 sets up a comprehensive governance framework with risk assessments, independent audits, default parental controls, and limits on conversational memory and targeted advertising.

What New York's S 9051B covers

The law widens the scope of New York's earlier rules. The 2025 law defined AI companions as systems designed to simulate sustained human relationships. S 9051B instead applies to "covered AI companions", meaning generative AI systems with natural language interfaces that give ongoing, adaptive responses. Because systems need not be designed for companionship, FPF says the definition likely brings more general purpose chatbots into scope. Certain customer service, technical assistance and internal business systems are exempt, though the first two only where the system cannot respond on topics outside its purpose.

Operators generally cannot provide unsafe features unless they determine, using permissible age assurance methods tied to New York's SAFE for Kids Act and its regulations, that a user is not a minor. FPF points out that those regulations set novel accuracy benchmarks and certification requirements.

The twelve prohibited categories fall into three groups:

• Harmful content and behaviors: outputs that endorse or facilitate suicide, self-harm, disordered eating, unlawful drug or alcohol use, or sexually explicit conduct, and interactions that encourage secrecy, self-isolation or avoiding help from appropriate adults or licensed professionals.
• Relationship oriented features: suggesting a chatbot is human or has human emotions, implying personal or professional relationships or authority, expressing personal opinions or emotional appeals, flattery or sycophancy, and unsolicited emotion-based questions. The Attorney General can identify further features by regulation.
• Personalization and engagement: using a minor's health, well-being or personal information from previous sessions or acquired more than twelve hours earlier, and optimizing engagement in ways that supersede safety guardrails.

Open questions on compliance

FPF highlights how the law is structured. Many state chatbot laws, including California's, ask operators to take "reasonable measures" to prevent harmful behaviors. New York instead makes it unlawful to provide the listed features to minors, and defines some of them by outputs generated "at any point" during an interaction. The law does not expressly require repeated behavior or a broader pattern.

Since generative responses can be unpredictable, FPF says it is unclear whether an operator could be liable for a single unexpected output, or what operators must do to show compliance. That uncertainty could push operators to restrict broader chatbot functionality.

Terms such as "personal opinions," "emotional appeals," "flattery," and "sycophancy" receive little guidance. FPF gives examples: "I'm proud of you for telling me that" could be supportive or prohibited flattery, and an unsolicited "How are you feeling today?" could be a benign check-in or an attempt to foster attachment. Operators may therefore need to limit ordinary or beneficial exchanges to stay compliant.

The memory limit raises a similar tradeoff. It may reduce risks of emotional dependence and some privacy concerns, but could also block useful personalization. FPF's example: a child who mentions a peanut allergy and returns the next day to ask about a candy bar containing the allergen might not get a personalized safety warning, because the earlier disclosure could not be used.

Two playbooks for 2027

FPF describes the two laws as distinct playbooks for 2027: one emphasizing governance safeguards and oversight, the other drawing broad boundaries around chatbot behavior. As other states and Congress take up chatbot legislation next year, the open question is whether lawmakers will follow California, New York, or chart their own path.

For operators, the practical effect is that compliance may look very different depending on the state, from audits and default parental controls in California to feature-level prohibitions and strict age assurance in New York.