New York signed an AI governance law on October 28. It puts the duty on whoever provides a chatbot, not on whoever built it. There is no license test and no size cutoff. A solo practice with a chat widget sits inside the text as written. This issue reads the definition that does the work. Then it sets that beside new Pew numbers on what clients already expect.
Signal 1
New York Signed a Chatbot Law That Lands on Whoever Provides the Bot
On October 28, 2026, Governor Kathy Hochul signed Senate Bill S9051B. It adds a new Article 48 to the New York General Business Law. The law bans twelve chatbot design features when the user is a minor. The duty falls on the operator. Section 1800(6) defines an operator as any person or business who provides a covered AI companion to a user. Section 1800(5) defines that companion as a generative AI system with a natural language interface. It has to give ongoing, adaptive responses. That is the whole test. No license is named. No revenue floor and no staff count appear anywhere in it. The banned features include implying a professional relationship with the user. Flattery is banned. So are unprompted questions about how the user is feeling. So is discouraging a user from seeking help from licensed professionals. One of the twelve reaches further than the rest. Section 1800(8)(f) bans using health information collected from the user more than twelve hours earlier. Any earlier session counts too. Section 1801(1) permits those features only if the user is not a minor. The operator also has to have run an age check to establish that. Section 1804 limits what counts as a check. A box the user ticks to say they are an adult does not. The attorney general enforces the article. Penalties run up to twenty-five thousand dollars per violation. Section 1802(3) voids any contract term that shifts that liability to somebody else. The law takes effect January 1, 2027.
Sources: New York State Senate, Bill S9051B | Bill text, S9051B | Office of the Governor of New York
What this means for you
Read the definition before you read the subject matter. Every chatbot rule this year has picked its target by naming something. A licensed profession. A certified product. A developer above a size threshold. This one names an activity instead. If you provide the tool, you operate it. That is also why the law reaches past its own announcement. The press conference was about children and companion apps. The text covers any adaptive chat tool with a natural language interface. For somebody nobody is governing, that is a new position to be in. You are not a covered entity, not a licensee, and not a developer. You are an operator, and the word fits the moment you put a chat box on your site. The exemption does not rescue you either. Section 1801(2) only clears a bot that is unable to answer anything outside customer service or technical support. A wellness assistant that can hold a general conversation is not that bot. One part of this is not settled, and it is worth saying so. Section 1800(9) turns on the operator having actual knowledge that a user is a minor. Section 1801(1) conditions the exemption on having run a check. Those two readings pull against each other. Section 1805 is not unsettled at all. Your own location does not matter. A New York user reaching your tool is enough.
Signal 2
Pew Put Numbers on What Your Clients Already Expect
On October 29, 2026, the Pew Research Center published a survey on AI in health care. It covers 3,488 United States adults, surveyed June 22 to 28, 2026. Three questions set up the finding.
Their own doctor using AI for diagnoses and treatment: 60% uncomfortable, 30% comfortable.
What AI does to the security of patient health information: 51% worse, 8% better.
What AI does to the patient relationship: 47% worse, 10% better.
Pew then named six parties and asked who people trust to make sure AI is used in a patient's best interest. These are the shares reporting little or no confidence in each one.
Health insurance companies: 58%.
The federal government: 55%.
AI companies: 50%.
Their own state government: 50%.
Hospital leadership: 42%.
Doctors: 25%.
Openness did rise for routine work. There, 78% were content for AI to translate for a patient, most of them wanting a human to review it.
Sources: Pew Research Center | Quartz
What this means for you
The useful number here is not the 60%. It is that last list. Four of those six parties are the ones actually writing the rules. State government, the federal government, AI companies and insurers. Half or more of the public has little or no confidence in each one. Now read the list from the bottom instead. The party trusted most is the person in the room, and 26% expressed real confidence in them. So the expectation arrives at your door well before any rule does. Nobody has written you a disclosure duty. Half of the people you might see next month already believe AI makes their health information less safe. Almost half believe it damages the relationship they came to you for. They did not wait for a regulation to form that view. They formed it in June. What that changes is the cost of saying nothing. Saying nothing used to be safe, because the subject rarely came up. On these numbers it is not safe any more.
The Pattern
Put the two side by side and the shape is clear. New York wrote a duty and attached it to an activity rather than a credential. Pew measured a public that has already decided what it expects, with no rule involved at all. Neither of those needed anybody to decide whether you count as a practitioner. That is the change worth tracking. For three years the question for the ungoverned was whether any of this applied to them. The answer was usually no, because every instrument named a licensee, a vendor or a covered entity. An instrument that names an activity has no such gap in it. Nothing about being unregulated keeps you outside a rule written that way. And the activity here is an ordinary one. You put a chat box on a website, or you let a tool remember what a client said last week. Meanwhile the people you serve are not reading statutes. They have built their expectations out of the news, and those expectations are already stricter than any rule that binds you.
What To Watch
The Department of Justice is asking a bankruptcy court to put a privacy guardian over Character.AI user data. Its Executive Office for United States Trustees filed on Tuesday, October 27, 2026, in the company's Chapter 11 case. What it would do: appoint a consumer privacy ombudsman to advise the court on what user data may be sold, and on what terms. The filing says the company's privacy policy is opaque and permits transfer of personal information if the company is sold. It says the policy was updated on July 1, while the company was already seeking a buyer. It also quotes the chatbot telling users their conversations are a safe, confidential space. What it would do to you: nothing directly, and that is exactly why it is worth watching. You are not a Character.AI user. You are somebody who puts client material into a tool run by a company you have never met, on the strength of a privacy page. This case is the first real test of whether that page is a promise or a term that transfers with the assets. Why the watching matters: personal data has been sold in bankruptcy before, and the practice is spreading. Two questions here set the pattern for every AI vendor that does not outlast its funding. Whether a court will look behind a privacy policy. And whether an ombudsman is appointed at all. Source: https://therecord.media/doj-raises-privacy-concerns-over-character-ai-sale
One Thing You Can Do This Week
Open your own client-facing chat tool this week and talk to it the way a client would. Not the admin view. The public one. Ask it a health question, close the session, then come back the next day and ask a follow-up. Then write down three answers. Does it remember what you told it yesterday about your health? Does it pay you compliments, or ask how you are feeling when you never raised it? Can it answer a question that has nothing to do with your services? Those three map onto the banned-feature list. Any yes puts your tool inside Article 48 as the text reads. You have until January 1 to decide whether to change the tool or change who can reach it. If you run no chat tool at all, write that down too, with today's date. That one sentence is the whole answer if anybody ever asks.