Legislation signed Wednesday by Gov. Kathy Hochul defines a dozen categories of 'unsafe AI companion features' that are prohibited whenever a companion chatbot interacts with a child under 18.

The banned behaviors include a chatbot suggesting it is a real person with emotions, opinions, or an actual relationship with the child, as well as messages that use flattery, involve sexually explicit material, or encourage secrecy or unhealthy behavior.

Violations can expose a developer to a lawsuit by the state attorney general's office, with penalties that could include monetary damages, shutting down the chatbot, and destroying data. The law takes effect January 1.

The law is notable for how far it reaches into product design rather than just disclosure.

According to Justine Gluck, a policy analyst at the Future of Privacy Forum, New York's statute underscores how quickly chatbot regulation is evolving, as states move beyond basic disclosures and crisis-response protocols to regulating the underlying design and operation of chatbots.

Jim Steyer, founder and CEO of watchdog group Common Sense Media, called it 'the strongest kids' AI safety bill in the country.'

The surprise is that a single state has moved to police the conversational mechanics of companion bots themselves, not merely require warnings, while Congress has done little to regulate AI.

With Congress doing little, state legislatures are packed with bills using consumer protection law to make AI more transparent and rein in chatbots and frontier models, with New York and California at the vanguard. In September, California Gov.

Gavin Newsom signed legislation requiring chatbot developers to assess a model's risk to children, show how they tried to eliminate those risks, and submit to an independent audit; he also banned addictive social media features for users under 16.

Meanwhile, eight states sued OpenAI earlier this month, alleging its chatbot violated state consumer protection laws by hurting children and deceiving adults, in what one activist called the most significant coordinated state action ever taken against an AI company.

Gluck cautioned that several categories in New York's law are broadly worded or undefined, raising questions about 'where ordinary conversational responsiveness ends and prohibited behavior begins,' which could complicate how the ban is enforced in practice.

She said the different regulatory approaches taken by California and New York could serve as models other states adopt after their legislatures reconvene next year.

The stakes extend beyond compliance costs: developers now face state attorney general litigation and potential product shutdowns, and governors acknowledge Washington has farther-reaching ability to clamp down on AI threats, even as President Trump has brushed off concerns and warned states not to regulate AI.

Companies that operate companion-style bots in New York will need to remove personhood claims, flattery-driven engagement, sexually explicit content, and prompts toward secrecy or unhealthy behavior from minor-facing interactions, or risk attorney general action that can include damages, shutdown of the chatbot, and destruction of data.

Expect other states to watch the rollout closely, with Gluck suggesting the New York and California templates could shape a wave of copycat bills when legislatures reconvene next year.