Who Does This Actually Protect?

I want to be upfront: this piece is a response to an article by John Coleman, published in his Substack newsletter, about New York's Senate Bill 7263. Coleman made arguments I found genuinely compelling, and I want to credit him for the framing. But I also want to be honest that reading his piece left me with more complexity, not less, because this is a topic where I have genuinely mixed feelings, and I think those mixed feelings are worth working through in public.

New York is moving toward passing a bill that would make AI chatbot operators legally liable for giving advice that, if given by a human, would require a professional license. Medicine, law, mental health, dentistry, social work, engineering, and eight more professions — fourteen in total. The bill, Senate Bill 7263, passed the Senate's Internet and Technology Committee unanimously in February 2026 and reached the Senate floor calendar shortly after. It is not yet law, but it is moving.

On the surface, the intent is reasonable. Nobody wants an AI chatbot pretending to be a licensed therapist with a vulnerable user who is in crisis. Nobody wants a patient making a serious medical decision based on an AI that presented itself as a doctor. Senator Kristen Gonzalez, who introduced the bill, framed it as consumer protection, a safeguard against the harm that comes when AI impersonates credentialed professionals.

That is where my own tension lives. I'm a free speech advocate. Restricting speech, even AI-generated speech, requires a very high bar. At the same time, I work at a law school and think about access to justice every day. I've seen what happens when vulnerable people receive confidently wrong information and act on it. Those two things are not always easy to hold together, and this bill forces me to try.

What the Bill Actually Does

SB 7263 prohibits chatbot operators from allowing their systems to provide any "substantive response" that would constitute the unauthorized practice of a licensed profession if given by a human. It creates a private right of action — meaning users can sue — for actual damages, plus attorneys' fees in cases of willful violation. And critically, it makes clear that providing a disclaimer that the user is talking to an AI is not enough to avoid liability.

The bill never defines what a "substantive response" means. There is no line drawn between factual information and professional advice. Is explaining what ibuprofen is factual? What about recommending a dose? Is explaining what an eviction notice is factual? What about explaining what options a tenant has in response to one? The bill does not say, which means chatbot companies will answer that question for themselves, and they will answer it conservatively, restricting anything that might expose them to a lawsuit.

The predictable result is not better AI advice. It is less AI engagement with any topic that touches a licensed profession — which, when you think about it, is a significant portion of the questions that people most urgently need answered.

Who Actually Gets Hurt

I work at a law school. I think about access to justice every day, and I have for a long time. The question I keep asking about this bill is a simple one: who is it actually protecting?

The people who can afford lawyers, doctors, and therapists are not meaningfully affected by what AI chatbots will or won't say. If their chatbot stops answering detailed medical questions, they call their doctor. If it won't explain a contract clause, they call their attorney. Is that annoying? Sure. Could it be a real hassle, depending on the situation? Absolutely. But even when it is a major inconvenience, it is still an inconvenience they can work around, because they have somewhere to go. That is a fundamentally different situation than having nowhere to go at all.

The people who cannot afford those professionals are in a different situation entirely. The tenant who gets an eviction notice at 10pm and types a question into their phone because they don't know what it means and have no one to ask. The person in a rural county triaging symptoms because the nearest doctor is an hour away and the copay is more than they can spare this week. The small business owner trying to understand what a contract clause means before signing something they can't easily undo. The first-generation college student trying to understand financial aid terms that nobody in their family has ever navigated.

For those people, the choice this bill creates is not AI advice versus a licensed professional. It is AI advice versus no advice at all. A chatbot that responds to every legal or medical question with "please consult a licensed professional" is not a safety feature. It is a wall. And it is a wall that only appears for people who cannot afford to go around it.

The Access to Justice Problem Is Already Real

This is not a hypothetical. New research released in March 2026 documented a sharp rise in people representing themselves in federal court, alongside growing evidence of AI use among those self-represented litigants. Federal courts impose demanding procedural requirements. The fact that people are navigating them without lawyers, and increasingly with AI assistance, tells you something about both the scale of the access gap and the role AI is beginning to fill.

Stanford's Justice AI Co-Pilots program is already helping legal aid organizations draft filings, navigate complex rules, and manage high-volume cases that would otherwise go unserved. These are not edge cases. They are the center of what access to justice actually looks like in this country for the majority of people who need it.

SB 7263, as written, would create significant liability risk for exactly these kinds of applications. The organizations and companies trying to build tools that help underserved people navigate legal, medical, and financial complexity would face the same liability exposure as a chatbot falsely claiming to be a licensed attorney. That is not a careful calibration. It is a blunt instrument.

What Good Regulation Would Look Like

I am not arguing against regulating AI in professional contexts. The case for thoughtful guardrails is real. An AI that presents itself as a licensed professional, that encourages users to forgo actual medical care, or that provides confident wrong answers to legal questions with serious consequences deserves serious regulatory attention.

But there is a meaningful difference between guardrails designed to protect people from harm and guardrails designed to protect incumbents from competition. The former would require clear disclosure that the user is interacting with AI, not a licensed professional. It would establish liability for AI that affirmatively misrepresents its credentials. It would require accuracy standards and audit mechanisms for high-stakes applications. It would draw a careful line between dangerous impersonation and useful information.

SB 7263 does some of this. It requires disclosure. It targets impersonation. But by leaving "substantive response" undefined and creating broad liability for anything that touches a licensed profession, it goes much further than its stated purpose requires, in ways that will predictably reduce access to information for the people who need it most.

The question worth asking before this bill moves further is not whether AI advice can be harmful. It can. The question is whether the harm this bill prevents is greater than the harm it causes, and who bears each of those costs. Right now, the answer to the second question is uncomfortable: the people most likely to be protected by the bill's intent are already the most protected. The people most likely to be hurt by its effects are the ones with the least protection to begin with.

That is worth naming, before a well-intentioned law makes a gap that already exists significantly wider.

I want to be clear about where I land, even if I land with some uncertainty. I don't think AI impersonating a licensed professional with a vulnerable person is a free speech issue. I think it's a harm issue. Preventing that kind of dangerous deception is legitimate and important. But this bill goes further than that, and the collateral damage of its overreach falls on the people who were already most underserved. That is the part I think deserves far more attention than it has gotten so far.

Coleman's piece helped sharpen where the line should be. The goal should be preventing dangerous impersonation and ensuring accuracy, not suppressing useful information exchange on behalf of incumbents who would prefer less competition. Those are different objectives. This bill, as written, doesn't always distinguish between them.