The Food and Drug Administration is confronting a regulatory gap that has been widening for years: millions of Americans now turn to generative-AI chatbots for mental health support, yet not a single one holds FDA clearance or approval as a treatment. Apps and chatbots marketed for mood tracking, journaling prompts, and crisis-adjacent conversation have proliferated on app stores largely as unregulated wellness products, a classification that has let them scale quickly but has also left users with little independent verification of their safety or effectiveness. The agency’s Digital Health Advisory Committee met on November 5, 2025 to begin sketching what oversight of these tools might look like, and the debate has only intensified through 2026 as nonprofit patient-safety group ECRI named AI chatbots the top health technology hazard of the year.
A Hypothetical Chatbot, a Very Real Problem
At the November meeting, committee members examined a hypothetical therapy device that mimics human therapist conversations, walking through scenarios spanning prescription and over-the-counter use, adults and adolescents, and diagnoses ranging from major depressive disorder to a broader mix of mental health conditions. The exercise reflected a core dilemma: FDA has authorized more than 1,200 AI-based digital devices for marketing overall, but none has been indicated specifically to address mental health, leaving a regulatory pathway that, in the words of UC Irvine psychological science professor Stephen Schueller, “isn’t sized right for them.”
Woebot’s Exit Signals the Cost of Waiting for Clearance
The stakes of that mismatch became concrete when Woebot Health retired its consumer chatbot app on June 30, 2025, after roughly 1.5 million people had used it. Founder Alison Darcy cited the expense and difficulty of pursuing FDA marketing authorization while large language models kept advancing faster than the regulatory framework could adapt — an early casualty of a category now expanding rapidly without a clear approval route.
Woebot’s shutdown was notable precisely because the company had positioned itself as one of the more cautious, clinically minded players in the space, having published peer-reviewed studies on its chatbot’s effectiveness for mild depression and anxiety symptoms. If a company with that track record concluded the regulatory path wasn’t viable, the episode raised doubts about whether smaller or less rigorous competitors would even attempt the same scrutiny.
Schueller and Jacobson Push for a New Pathway
Writing in JAMA in a piece published January 9, 2026, Schueller and colleague Nicholas Jacobson argued that the millions of people already relying on AI chatbots for mental health support represent a novel product category that traditional FDA approval pathways were never built to evaluate, and that regulators need frameworks tailored to conversational AI rather than retrofitted device rules designed for static software or hardware.
ECRI’s Warning: Misleading Answers, Real Harm
Patient-safety organization ECRI’s decision to rank AI chatbots as the single top health technology hazard of 2026 underscores the risk regulators are racing to catch up with: chatbots can offer false or misleading information that could harm patients, particularly in mental health contexts where a wrong or poorly timed response carries outsized consequences. The FDA’s own advisory discussion noted the same core concern, describing generative-AI outputs as “unpredictable” and capable of “misleading users or causing patient harm.”
Industry Groups See a Difficult Line to Walk
Not everyone treats the hazard framing as settled fact. Companies building AI mental health tools argue that unregulated access to conversational support fills a real gap for people who cannot afford or access a human therapist, and that overly cautious rules could push those users back toward getting no support at all rather than toward a licensed clinician. Legal and regulatory analysts, including commentary from law firm Sidley Austin tracking FDA and CMS actions on generative-AI-enabled mental health devices, have noted that the agency’s own posture on AI regulation had been notably reticent for the past several years, making this new round of advisory committee attention a meaningful, if overdue, shift.
Some clinicians go further, arguing that any framework modeled on traditional medical-device review — built around a fixed version of software that behaves predictably — will always lag behind large language models that are retrained and updated on a rolling basis, meaning a chatbot cleared today could behave differently within months of an underlying model update.
What’s Next
The FDA has not set a timeline for issuing final guidance, and the committee’s November session was explicitly framed as fact-gathering rather than a step toward specific recommendations. In the meantime, wearable and app makers — from Whoop’s newly launched clinician video consultations to Oura’s virtual-care partnership with Counsel Health — are moving mental and physical health guidance further into AI-mediated territory even as the core question of how to certify a therapy chatbot as safe and effective remains unresolved.
Photo: DariuszSankowski / PIXABAY via Pixabay
