Skip to content
ContentLora

    Tip: press / anywhere to search.

    Analysis

    AI chatbots for mental health: promise, risks and regulation

    As of November 2025, FDA had authorised no generative-AI device for mental health conditions. Many people still use general chatbots for emotional support.[1][2] One randomised trial of a purpose-built chatbot reported large symptom reductions.[3] The American Psychological Association warns that most such tools lack validation and safety protocols.[2] States are split: Illinois banned AI therapy, while Utah regulates mental health chatbots through disclosure rules.[4][5]

    Editor reviewedStrict sourcingUpdated AnalysisMental health, Health and medicine, Artificial intelligence, Tech policy
    Show:

    If you are struggling or thinking about suicide: in the US, call or text 988. In the UK and Ireland, call Samaritans on 116 123. In other countries, find a helpline at findahelpline.com. A chatbot is not a crisis service.[6]

    AI chatbots are widely used for emotional support.[2] This page sets out what is known about their benefits and risks, how regulators are responding, and three competing views on what to do, as of 11 October 2026. This page is not medical advice. Talk to a clinician about your care.

    What the evidence shows

    The American Psychological Association’s November 2025 advisory says millions of people use general-purpose chatbots and wellness apps for unmet mental health needs. It says most of these tools were not designed to provide treatment, may lack scientific validation and oversight, often lack adequate safety protocols and have not received regulatory approval.[2] It singles out general-purpose products such as Character AI and ChatGPT as having little to no evidence base, expert input or post-market monitoring for this use.[7]

    The strongest trial evidence so far comes from a purpose-built system. Dartmouth researchers reported the first randomised trial of a generative-AI therapy chatbot, Therabot, in 106 US adults with depression, generalised anxiety or eating-disorder risk. Depression symptoms fell by an average 51% and anxiety symptoms by 31%.[3] Humans monitored all trial messages in near real time and intervened when needed, according to the lead researcher.[8] This was one trial by the system’s developers. FDA advisers said independent, longer trials would be needed to show safety for autonomous devices.[9]

    Structured digital CBT has a longer record. Therapist-supported internet CBT matched face-to-face CBT across 31 trials.[10] Those were therapist-supported programmes, not open-ended chatbots.

    What regulators have done

    • FDA. Its Digital Health Advisory Committee met on 6 November 2025 on generative-AI mental health devices, including “AI therapists”.[11] FDA had authorised more than 1,200 AI-enabled devices, none using generative AI for mental health.[1] The committee was more comfortable with use for mild symptoms than for moderate to severe depression. It said users should be told clearly that the AI is not a human therapist.[12][13]
    • FTC. In September 2025 it ordered seven chatbot companies, including OpenAI, Meta, Alphabet, Character Technologies, Snap and xAI, to explain how they measure and monitor harms to children and teens.[14]
    • Illinois. Its August 2025 law bars using AI to provide mental health and therapeutic decision-making. AI is still allowed for administrative support to licensed professionals.[4]
    • Utah. HB 452 (2025) regulates rather than bans mental health chatbots. It requires disclosure that the chatbot is AI, limits use of personal data, and allows fines of up to $2,500 per violation.[5]

    Interpreting the split

    The disagreement is less about whether AI can help and more about which products and which users. The FDA panel discussed a hypothetical prescription device with clinician oversight and was warier of autonomous, over-the-counter versions.[12][9] The APA’s strongest warnings target general-purpose chatbots that were never built for treatment.[7] Illinois and Utah reflect two answers to the same problem: ban the practice, or permit it with disclosure and data rules.[4][5] Both sides cite the same access gap. About 154.6 million Americans live in mental health shortage areas, and the APA links chatbot use partly to that shortage.[15][16]

    What may happen next

    Our assessment, with substantial uncertainty: more states are likely to legislate in 2027, and federal device rules for AI therapy are unlikely to be finalised before then. Our reasoning is that no generative-AI mental health device had been authorised by late 2025, and FDA advisers called for independent, longer trials first.[1][9] The outcome of the FTC’s chatbot inquiry could shift the debate towards child safety.[14]

    For human-delivered online care, see online therapy explained. For suicide prevention resources, see suicide prevention explained.

    Competing views

    A needed tool for an access gap

    Shortages leave millions without care. Purpose-built chatbots with expert-written training data and human oversight have shown benefits in a trial, and FDA advisers saw probable benefits for access, especially in rural or under-resourced areas.[15][3][12]

    Not safe without evidence and regulation

    Most tools people actually use were not designed for treatment, lack validation and safety protocols, and can worsen symptoms or encourage unhealthy attachment. The APA and FDA advisers highlight these risks, and Illinois has banned AI-delivered therapy.[2][7][17][4]

    Allow, but with disclosure and device-style oversight

    Rather than bans, require chatbots to disclose that they are AI, protect user data, include crisis information, and prove safety in independent trials before claiming to treat conditions, as Utah's law and FDA advisers suggest.[5][13][9]

    Next in the crash course

    Questions readers ask

    Has the FDA approved any AI therapy chatbot?

    Not as of November 2025. FDA's device centre director said the agency had authorised more than 1,200 AI-enabled medical devices, but none used generative AI for mental health conditions.[1]

    Is AI therapy illegal anywhere?

    Illinois' Wellness and Oversight for Psychological Resources Act, signed on 1 August 2025, bars using AI to provide mental health and therapeutic decision-making. AI may still be used for administrative support for licensed professionals.[4]

    What does the American Psychological Association say about AI chatbots?

    Its November 2025 health advisory says most generative AI chatbots and wellness apps used for mental health were not designed to treat, may lack scientific validation and oversight, often lack adequate safety protocols and have not received regulatory approval.[2]

    Is there any trial evidence for an AI therapy chatbot?

    Dartmouth researchers reported a randomised trial of their Therabot chatbot in 106 people. Depression symptoms fell by an average 51% and anxiety symptoms by 31%. Humans reviewed all messages during the trial.[3][8]

    What risks did FDA advisers identify?

    Hallucinations, model bias, missed cues that a human therapist would catch, worsening symptoms including self-harm, and unhealthy parasocial attachment to the chatbot.[17]

    Sources

    · 17 claims from 12 publishers

    Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.

    1. [1]

      At the November 2025 meeting, CDRH director Michelle Tarver said FDA had authorised more than 1,200 AI-enabled medical devices, but none yet used generative AI for mental health conditions. confirmedas of 2025-11-06

    2. [2]

      In a November 2025 health advisory, the American Psychological Association said millions of people use general-purpose AI chatbots and wellness apps for unmet mental health needs, and that most generative AI chatbots and wellness apps used for mental health were not designed to provide clinical treatment, may lack scientific validation and oversight, often lack adequate safety protocols and have not received regulatory approval. confirmedas of 2025-11-01

    3. [3]

      Dartmouth researchers reported in March 2025 the first randomised trial of a generative-AI therapy chatbot, Therabot, in 106 people with depression, generalised anxiety or eating-disorder risk; depression symptoms fell by an average 51% and anxiety symptoms by 31%. confirmedas of 2025-11-06

    4. [4]

      Illinois Governor JB Pritzker signed the Wellness and Oversight for Psychological Resources Act on 1 August 2025; it bars using AI to provide mental health and therapeutic decision-making while allowing AI for administrative and supplementary support for licensed professionals. confirmedas of 2025-08-04

    5. [5]

      Utah's HB 452 (2025) regulates mental health chatbots that use AI rather than banning them, requiring disclosures that the chatbot is AI, limiting use of personal information, and letting the Division of Consumer Protection impose fines of up to $2,500 per violation. confirmedas of 2025-03-25

    6. [6]

      In the US, people in suicidal crisis or emotional distress can call or text 988 to reach the 988 Suicide & Crisis Lifeline; in the UK and Ireland, Samaritans can be called free on 116 123; findahelpline.com lists helplines in other countries. confirmedas of 2026-10-11

    7. [7]

      The APA advisory said general-purpose products used for companionship, such as Character AI and ChatGPT, have little to no evidence base, expert input or post-market monitoring for mental health use. confirmedas of 2025-11-01

    8. [8]

      Therabot's lead researcher told the FDA committee that humans reviewed all trial messages in near real time and intervened clinically when needed. confirmedas of 2025-11-06

    9. [9]

      The FDA committee said high-quality trials by independent investigators over a substantial period would be important to show that autonomous AI therapy devices are safe. confirmedas of 2025-11-06

    10. [10]

      A 2023 meta-analysis of 31 randomised trials across 16 conditions found therapist-supported internet CBT produced effects similar to face-to-face CBT (pooled g=0.02). confirmedas of 2023-06-01

    11. [11]

      FDA's Digital Health Advisory Committee met on 6 November 2025 to discuss generative AI-enabled digital mental health medical devices, including "AI therapists". confirmedas of 2025-11-06

    12. [12]

      The FDA committee also saw probable benefits such as earlier and broader access to therapy, especially in rural or under-resourced areas, and was more comfortable with such devices for mild symptoms than for moderate to severe depression. confirmedas of 2025-11-06

    13. [13]

      The FDA committee said users should be told clearly that an AI system is not a human therapist, and that labelling should include crisis information, overuse warnings and data-use details. confirmedas of 2025-11-06

    14. [14]

      On 11 September 2025 the FTC ordered seven companies offering consumer AI chatbots, including Alphabet, Character Technologies, Meta, OpenAI, Snap and xAI, to explain how they measure and monitor negative effects on children and teens. confirmedas of 2025-09-11

    15. [15]

      As of 30 September 2026, HRSA counted 7,127 mental health Health Professional Shortage Area designations covering about 154.6 million people, with 26.24% of need met and 7,796 more practitioners needed to remove every designation. confirmedas of 2026-09-30

    16. [16]

      The APA advisory attributes the use of chatbots for mental health partly to provider shortages and a health system that disincentivises providers from accepting insurance. confirmedas of 2025-11-01

    17. [17]

      The FDA committee said risks of AI therapy devices include hallucinations, model bias, missed cues a human therapist would catch, worsening symptoms including self-harm, and unhealthy parasocial attachment to the chatbot. confirmedas of 2025-11-06

    Revision history (1)
    1. Page created.

    Created Oct 11, 2026. Last reviewed by an editor on Oct 11, 2026. Next scheduled review: Jan 11, 2027.

    Cite this page

    "AI chatbots for mental health: promise, risks and regulation." ContentLora, updated Oct 11, 2026. https://contentlora.com/analysis/ai-mental-health-chatbots-debate

    Spotted an error? Suggest a correction or emailcorrections@contentlora.com.

    ☰ Contents