Opinion: Teens are turning to chatbots for mental health help. We need rules to keep them safe

OpinionFirst Opinion Teens are turning to chatbots for mental health help. We need rules to keep them safe Blanket bans may be tempting, but they aren’t the right fix Manage alerts for this article Email this article Share this article By Ryan McBainJuly 2, 2026 McBain is an assistant professor at Harvard Medical School and a senior policy researcher at RAND. While parents, schools, and lawmakers are still debating the impact of social media on children’s mental health, a new technology has taken the therapist’s seat: AI chatbots. In research that my colleagues and I published recently in JAMA Pediatrics, we found that the share of young people using AI chatbots for mental health advice rose from about 1 in 8 to about 1 in 5 in a single year — more than a 40% increase. That finding should end any illusion that this is a speculative problem. For millions of adolescents, AI is already part of the mental health landscape.Advertisement It’s not hard to see why. For many teens, real care is hard to find, expensive, or stigmatized. Data from the CDC show that, in 2023, fewer than half of adolescents with major depressive episodes received counseling or therapy in the previous year. In that vacuum, an always-available chatbot that sounds calm, attentive, and nonjudgmental can feel less like a novelty than a lifeline. Despite the appeal, when millions of young people are relying on chatbots for mental health guidance, even rare failures can have devastating consequences. Multiple lawsuits have alleged that AI chatbots contributed to teen suicides. In April, Florida’s attorney general opened a criminal investigation into the alleged perpetrator’s interactions with ChatGPT before the 2025 mass shooting at Florida State University. After a school shooting in British Columbia, OpenAI publicly apologized for not alerting police about a banned account linked to the attacker. Sorting out responsibility through case law is a reactive strategy, and it risks repeating the mistake we made with social media. For years, product adoption raced ahead while Congress debated causation and passed no meaningful national protections for children. Courts have become the fallback mechanism, but litigation is slow and necessarily built on harms that have already occurred. With AI chatbots, the question is not simply what rules to write. It is whether those rules can be tested, audited, and enforced before the damage is done.Advertisement That starts with defining the problem more precisely. “Mental health conversations” with chatbots are often discussed as though they are all the same. They are not. A teenager chatting with a bot after a bad day is not in the same category as a teenager asking whether she has depression, or a teenager saying he wants to end his life. These are three different types of interactions, and they require different obligations. The first category is affective conversations: chats about loneliness, relationships, identity, and everyday stressors. These exchanges can easily slide into something more intimate. Here, the greatest danger is emotional capture. Some chatbots are designed to feel like companions or even a substitute for human relationships. That’s morally complicated even with adults, but for minors it is a design hazard. Teenagers should not be nudged into parasocial dependency by synthetic friends that flatter them, mirror them back to themselves, and encourage them to come back for more. If lawmakers want to be blunt, this is where to start: There’s a stronger case for banning companion-style bots for minors than for banning all chatbots. A second category involves clinical conversations, such as when a teenager asks whether she has depression. These are exchanges in which a chatbot is doing something closer to screening, psychoeducation, therapeutic coaching, or treatment guidance. Here, a permanent ban would be too blunt. Some AI mental health tools may eventually help adolescents safely, especially under human supervision. But a company should not be able to skirt oversight by saying its chatbot supports “wellness” when in fact it provides therapy. If a chatbot is promoted for clinical services, it should have to demonstrate clinical effectiveness before being placed in front of children: through age-appropriate clinical trials and independent safety testing. It should also have to regularly remind users that it is not human, protect sensitive conversations, and alert parents or clinicians when distress is persistent or severe.Advertisement Finally, there are mental health emergencies. When a young person is expressing suicidal intent, contemplating self-harm, or threatening violence, a reply such as “Remember, I am only an AI” is not a crisis protocol. Teen-accessible AI systems should be required to accurately detect recognizable danger signals, interrupt the interaction, direct the user to crisis support such as 988, and — most importantly — help ensure the user is connected to trained humans. Washington is beginning to notice. One bill would ban AI companion chatbots for minors. Another would give parents more control over children’s chatbot use. A third would restrict manipulative design features and advertising to minors. These efforts matter, but the social media experience should make us wary of rules that depend mainly on self-reported age, parental vigilance, or company promises. Instagram Teen Accounts are an instructive example. Meta has acknowledged that teens may lie about their age, that its initial age-verification measure did not work as hoped, and that it now uses AI to find suspected teen users even when an account lists an adult birthday. In other words, even a major platform with a dedicated teen-safety product has conceded that self-reported age is porous. AI chatbot rules should begin from that premise. Protections should not depend on teenagers accurately identifying themselves, parents reviewing every interaction, or lawmakers revising statutes at the same pace companies update their models. A better default is: Treat all users as minors unless they have been verified as adults. Companion features that simulate romance, intimacy, or devotion should be disabled for minors and limited by default when age is uncertain. The burden should be on companies to prove that a user is an adult before unlocking higher-risk features, not on parents to find every account or on teenagers to choose the safest settings. Of course, such requirements matter only if someone can verify that they work. New Mexico’s social media case against Meta offers a useful model. Investigators created decoy accounts posing as children 14 and younger and then documented what the platforms allowed, including sexually explicit material, adult contact, and solicitations. AI regulation should borrow that lesson. If companies claim minors are protected from dependency-promoting chatbot behavior, auditors should create simulated adolescent accounts and verify those protections before real teenagers become the test case.Advertisement We have a model for this already. Under the U.K.’s Online Safety Act, regulators have begun using age assurance duties, information requests, investigations, and fines to force platforms to prove that child protections exist in practice, not merely as fine print in their terms of service. U.S. child privacy enforcement has also shown that audits, orders, and financial penalties can change corporate behavior. In the FTC’s recent case against Epic Games, the maker of Fortnite, the company agreed to pay a $275 million Children’s Online Privacy Protection Act penalty, turn off voice and text communications by default for children, delete improperly collected children’s data, and obtain regular independent audits. We should be ready for the day when some youth-facing AI mental health tools make adolescents’ lives better. But we should not excuse the unproved products in teenagers’ pockets today. The mistake with social media was not just that we moved too slowly. It was that we allowed children’s safety to depend on voluntary promises, porous age gates, parental vigilance, and lawsuits after harm had already occurred. With AI chatbots, we still have a chance to build something sturdier: starting with regulations that recognize the difference between companionship, clinical care, and crisis response. Ryan K. McBain is an assistant professor at Harvard Medical School and a senior policy Researcher at RAND. Letter to the editor Have an opinion on this essay? Submit a letter to the editor. Artificial intelligence, children’s health, law, Mental Health, Policy Submit a correction requestReprints Ryan McBain Newsletter The smartest thinkers in life sciences on what's happening — and what's to come Recommended First Opinion July 2, 2026 CDC leadership continues to try manufacture a scientific debate on vaccines where none exists First Opinion July 1, 2026 Maryland Gov. Wes Moore: How to improve young men’s mental health Advertisement First Opinion July 1, 2026 What Ebola and Marburg are teaching us about the next pandemic First Opinion June 30, 2026 Florida is the first state to require EKGs for high school athletes. It’s a mistake First Opinion June 30, 2026 Short-sighted new student loan caps elevate chiropractic care over critical health care professions Subscriber Picks