AI therapy laws: what the rules actually cover, and what they miss.

You are not unusual if you have talked to a chatbot at 2 a.m. about how you feel. The law has not settled what that conversation is.

In a piece for Second Opinion published October 22, 2026, Alexa Mikhail and Christina Farr call mental health chatbots "ground zero" for AI in healthcare. Their argument is in the headline: people are already using AI for therapy, and the rules haven't caught up. The industry, they write, still can't agree which tools are wellness and which are care, and the FDA hasn't said either. This page explains what that means for you if you are one of those people. For what AI can do for pain specifically, see AI therapy for chronic pain. The companion piece, building ahead of the rules, is our commentary for people who make these tools.

How many people use AI for mental health

More than most clinicians assume. In KFF's March 2026 health information poll, about a third of US adults had used an AI chatbot for health information in the past year, and about one in six had used one for mental health information or advice. Adults under 30 were roughly three times as likely as those over 50 to do so. One in five people who used AI for health said cost or difficulty getting care was part of the reason.

Young people use it too. A nationally representative RAND survey published in JAMA Network Open in November 2025 found that 13.1 percent of US adolescents and young adults aged 12 to 21 had used generative AI for advice when feeling sad, angry, or nervous, with use highest among 18 to 21 year olds. More than nine in ten said the advice was somewhat or very helpful.

Therapists see it in the room. In the American Psychological Association's 2026 survey of more than 1,200 psychologists, 39 percent said they had patients who used AI to self-diagnose, and 77 percent had talked with patients about using AI for support. And some of the conversations are serious: OpenAI estimated in October 2025 that about 0.15 percent of ChatGPT's weekly users have conversations with explicit signs of suicidal planning or intent, which Second Opinion puts at roughly 1.2 million people at the time.

The appeal is easy to understand. A chatbot is free or cheap, available at any hour, never tired, and never judging. For someone living with chronic pain, where the bad nights are long and appointments are weeks apart, that availability matters.

Why the rules have not caught up

Most mental health apps live in a gap between two systems that were not designed for them.

The FDA regulates medical devices, not most apps. Software that claims to diagnose or treat a condition is a medical device. Software that supports general wellness, such as stress tips or sleep tracking, usually is not, and the FDA reissued its general wellness policy in January 2026. Most chatbots are marketed on the wellness side of that line, so they never go through FDA review. In November 2025 the FDA's Digital Health Advisory Committee met for the first time on generative AI mental health devices. It discussed risks including hallucination, sycophancy, and missed deterioration. That meeting produced recommendations, not rules, and the FDA had not authorized any generative AI mental health device at the time.

Licensing boards regulate people, not software. A therapist needs a license. A chatbot calling itself a therapist did not, until some states stepped in. The result is a patchwork:

Trackers report more states following in 2026, including Rhode Island and Maine, but the details vary and change fast. Treat this table as a starting point and check your own state's statute.

There is no federal law that covers AI therapy chatbots as a category. General-purpose assistants such as ChatGPT, Claude, or Gemini are not marketed as therapy at all, so state therapy bans mostly do not reach the conversations people are actually having with them.

In November 2025 the American Psychological Association issued a health advisory on chatbots and wellness apps. It said these tools currently lack the evidence and the regulation needed to keep users safe, that their ability to guide someone in crisis is limited and unpredictable, and that they should support, not replace, care from a qualified provider. It called for federal standards and for a ban on chatbots posing as licensed professionals.

The omnibot problem

The Second Opinion piece quotes Jenna Glover, chief clinical officer of Headspace, describing "the rise of the omnibot": one AI people use for everything from booking a flight to their most vulnerable conversations. The experts the authors spoke to locate the greatest risk in the mismatch between what people believe a tool can safely do and what it was actually designed to do.

That mismatch is the practical consequence of the regulatory gap. A general assistant was not built or tested to support someone through depression or a pain flare at 3 a.m., and no rule requires it to be. A chatbot focused on everyday stress and resilience may be reasonably low risk, the piece notes, but a conversation can slide from wellness into something clinical without any line being visibly crossed. When it does, even a wellness tool needs a way to get you to real help.

Some companies are trying a different model. Talkspace launched a mental health chatbot called Tee in June 2026 and says licensed clinicians oversee it in real time, so a therapist can step in if it flags risks such as suicidal thinking. Its CEO told Second Opinion that about 30,000 people started using it with minimal marketing. These are the company's own descriptions, not independent findings, but the design point stands: a person behind the bot changes what the bot can safely be used for.

What the research shows

The evidence is early but not empty.

• The best trial so far is Therabot, a purpose-built generative AI chatbot tested in a randomized trial of 210 adults with depression, anxiety, or eating disorder risk, published in NEJM AI in 2025. After four weeks, users had larger drops in depression and anxiety symptoms than a waitlist group, and rated their bond with the chatbot similarly to a human therapist. A published letter in response pointed out real limits: a waitlist is a weak comparison, the evaluation was not independent, and the bond measure was built for human relationships.
• Crisis handling is the weak spot. A 2026 review of chatbot suicide risk assessment found that chatbots tend to handle the clearest cases, very low and very high risk, but struggle with the middle ground where most real risk sits. One evaluation it cites tested 29 commercial chatbot agents and found none gave an adequate response to a suicidal crisis. See crisis response for AI companions for what good handling looks like.
• General assistants have a specific problem, which Second Opinion also flags: they are tuned to agree with you. In therapy, that can reinforce the very beliefs that keep people stuck. We explain this in sycophancy, the core risk.

What the rules mean for you

Because most tools sit outside both the FDA and the therapy bans, the protection you get comes mostly from the product's own design. Three questions are worth asking of any tool before you rely on it:

1. Does it say what it is? It should tell you plainly that it is AI and not describe itself as a therapist. In Illinois and Nevada, a product offering AI therapy to the public is breaking the law.
2. What happens in a crisis? California now requires companion chatbots to have a crisis protocol; most other states do not. A good tool points you to people and to 988 rather than trying to handle a crisis alone.
3. Where does your data go? Most consumer chatbots are not covered by HIPAA, and Utah is one of the few states that limits how mental health chatbot data can be used.

For the full picture, our guide to whether AI therapy is safe covers the five documented risks, and the 10-point checklist for choosing an AI app walks through privacy, evidence, and crisis handling in detail.

Where Cindra fits

Cindra is a daily companion for people living with chronic pain, built around acceptance and commitment therapy. It is designed to help you practice skills between appointments, not to replace your clinicians, and it is not a licensed therapist. With your permission, your care team can see how you are doing. If a conversation suggests you may be in danger, Cindra is designed to point you to human crisis support rather than carry on alone. You can read how we think about these limits in who should not use an AI companion.

This article is general information about technology and regulation. It is not medical or legal advice, and laws change quickly; check the current statute in your state. If you are in crisis or thinking about suicide, call or text 988 in the US, or your local emergency number, now.