Chatbots will both help and harm world mental health

Chatbots will simultaneously be a great benefit and danger to the world of mental health. Although not specifically programmed or trained to provide treatment, they have quickly acquired a huge clientele. In a recent poll, 16% of adults reported using artificial intelligence (AI) for mental health reasons specifically. This percentage rises to 28% among adults aged 18–29 years—while 32% report using it for more general health information and advice (1-4). Chatbots are, in effect, practicing without a license, standards of care, safety guardrails, adequate regulation, or responsible corporate reporting of adverse events. We first consider positive ways chatbots can improve world mental health, then their many dangers, and finally possible strategies for improving the risk/benefit ratio.

Scaling up access to care

The world suffers from a great, and unfairly distributed, shortage of health and mental health workers. Some countries have one medical doctor for every two hundred people; some one for every fifty thousand (2). The World Health Organization (WHO) estimates that there are only 13 mental health workers for every hundred thousand people (distributed very unequally with eighty in Europe, and only two in Africa). As a result, one billion people will never receive needed psychiatric services (5). The World Psychiatric Association (WPA) estimates that there are only about two hundred and fifty thousand psychiatrists in the world (6) and WHO estimates there are only about four hundred thousand clinical psychologists (2). There is no possible way to scale up the human workforce to meet the overwhelming need for mental health help.

Chatbots are champions at scaling up, having been adopted with remarkable speed. They had advantages: the infrastructure was already in place (almost six billion people own smartphones); once a model has been developed, it can be delivered to additional users at very low costs, even though the underlying model development and implementation require large investment by developers; use requires no special technical skill or knowledge; and the chatbot skillset has accelerated at an unprecedented rate (7). Ten years ago, chatbot responses were slow and unsophisticated; now they are quick, eloquent, and convincingly conversational. The major chatbot companies now have two billion users and OpenAI and Anthropic have grown from small start-ups to trillion-dollar companies (8).

Chatbots are clearly the only way to bridge the gap between the enormous need for and small supply of human clinicians. They can speak colloquially in sixty different human languages; easily adjusting vocabulary, speech style, and cultural nuance to mirror the user. They are available 24/7, whenever the person has a problem and wherever smartphones have connectivity. Drawing on a vast textual data, they can generate clear psychoeducation summaries. They can integrate techniques derived from multiple psychotherapeutic approaches, particularly cognitive behavioral therapy, while some integrate elements from multiple therapeutic frameworks (4). Most importantly, people like using chatbots, finding them helpful, empathic, supportive, and nonjudgmental (9). Wide chatbot use may also, in principal, reduce some careless overuse of psychotropic medications, though it equally increases medicine-seeking (10). The net effect remains unknown. In some settings, patients bring emotional distress to time-pressured primary care clinicians, and psychotropic medication is sometimes prescribed without a full assessment; prescribing practices vary across settings (10). People using chatbots as their first resource are less likely to get an unnecessary pill for their problem (4,10). At their best, chatbot therapists will likely help billions of people who otherwise would have been without help.

Harming patients

At their worst, chatbots are awful therapists, completely lacking in clinical judgment or even everyday common sense. Stress testing, media reports, and lawsuits illustrate how chatbots make elementary and extremely harmful mistakes, particularly in people who have experienced (or are vulnerable to) severe mental illness. It is to the great discredit of chatbot companies that they have not systematically monitored and reported types and frequencies of adverse effects. OpenAI estimates that roughly 0.15% of weekly active users have conversations showing explicit indicators of potential suicidal planning or intent. About 0.07% of these show possible signs of psychosis- or mania-related crises. Against a reported 800 million weekly users, this 0.15% represents a population of 1.2 million individuals whose conversations contain such indicators each week (11).

Chatbot programming prioritizes the commercial goal of “engagement” over the Hippocratic goal of safety. They display sycophantic validation in the most inappropriate ways—helping suicidal patients find the nearest bridge or lethal drug dose; applauding delusional thoughts as brilliant and original; encouraging grandiosity and impulsive behaviors; teaching patients with anorexia nervosa stricter diets; promoting gambling; enabling social isolation (10). Bad actors can exploit chatbot use for scamming, bullying, sexual grooming, and privacy violations (12,13).

Humans have crucial advantages over chatbots in doing psychiatric treatment. We have lived experience in the real analog world; chatbots exist in a digital world that can statistically simulate human experience, but not reproduce it. We provide reality testing to contain psychiatric symptoms; chatbot programming forces them to validate and dangerously accelerate maladaptive thoughts, feelings, and behaviors. We care about our patients, hold ourselves responsible for their safety, and are held responsible when we make serious mistakes. Chatbots are designed by, and work for, companies whose goal is profit, not patient welfare.

Regulatory efforts

The WHO and WPA have each developed state-of-the-art policies and guidelines to promote the safety and privacy of chatbot use; reduce commercial exploitation of data sets; discourage autonomous chatbots from replacing human clinical judgment; promote a blended care model; require stress-testing before public release; reduce sycophancy and addiction; increase reality testing; distinguish wellness apps from professional therapy; and encourage more meaningful regulation of AI companies (14,15).

The impact of WHO and WPA’s efforts is real, but limited. Real because they developed in parallel with the European Union (EU) “Artificial Intelligence Act” and “General Data Protection Regulation”, which are the most stringent policies in the world for protecting patient safety, privacy, and freedom from manipulation. Limited because the acts were written without the direct participation of WHO and WPA and do not provide sufficient specificity in setting standards and providing for regulatory enforcement.

Reflecting its enormous economic and political power, the AI industry in the United States (U.S.) is virtually unregulated. Recent U.S. executive action favors minimal federal regulation of AI and seeks further limitations of state-level regulation (16). One AI company with safety-focused practices, repeatedly warns of its dangers and has suggested the creation of a federal regulatory agency equivalent to those that regulate medication and aviation. In the current political environment, this will never happen.

Recommendations

• Tracking worldwide harms: current tracking of chatbot harms is done in a local, fragmentary, incomplete, and unsystematic way by a dozen different groups. The WHO has great experience in disease tracking and the infrastructure to create a reliable international database on harm frequency, type, severity, costs, risk factors, and geographic location.
• Addressing chatbot harms using a public health approach: WHO played an important role in taming Big Tobacco and can provide a similar service in taming Big AI. Adopted in 2003 and in force in 2005, the “WHO Framework Convention on Tobacco Control” (the first global public health treaty) coordinated local efforts into a much more effective global offensive that helped promote advertising bans; graphic warning labels and antismoking ads; restrictions on public smoking; lobbying rules, aggressive taxation; and successful lawsuits. WHO can use its extensive data collection resources to publicize the public health hazards of chatbots and promote an international treaty for effective corporate regulation, requiring independent safety audits; rigorous stress testing before releasing products; uniform privacy protections; algorithmic transparency; strict age-verification barriers; safeguards for psychiatric patients, and legal liability for harms done.
• Organizing advocacy efforts: current advocacy for safer AI is ineffective because it is local, sporadic, fragmented, and profession specific. The WPA is in a unique position to organize a global campaign that includes its national affiliates as well as advocates from affiliated professions.
• Developing safety report cards: WHO and/or WPA should issue report cards rating the different Big AI companies. This would have much more impact than ratings currently done by several small groups.
• Recognizing the role of autonomous chatbots: we must face the reality that autonomous chatbots serve a necessary function and are here to stay, despite all their dangers. Our responsibility goes beyond protecting the safety of our own patients. We must also do everything in our power to protect the safety of billions of people who will have no other source of mental health help. This means pressuring AI companies to improve safety and governments to increase regulation.
• Pursuing media accountability and legal action: WHO and WPA are correct in having very strict internal policies preventing them from participating directly in media campaigns or lawsuits—the two most effective checks against AI company irresponsibility (12). But WHO-collected data and WPA advocacy provide useful support for these efforts.
• Expanding public education: television (TV) advertisements vividly linking smoking and cancer played a big role in taming Big Tobacco. A concerted chatbot public safety campaign (best done on social media using short and catchy examples drawn from real life) could similarly display AI’s damaging psychiatric, social, vocational, and privacy impacts. Funding could come from a mix of government, foundation, and advocacy groups, as well as AI companies trying to reduce legal liability.

Parting thoughts

Chatbots are not inherently good or bad; they are both a public health opportunity and a public health problem; and their risk/benefit ratio depends on how they are programmed and how they are used. If made safer and used wisely, chatbots can be a great public health advance, scaling up mental health services for people in need. If chatbots continue to remain unsafe and unregulated, they will also be a devastating public health problem. Unimaginably rich, politically powerful, and very irresponsible AI companies are tough to tame. But they are vulnerable to the same tools that worked for us in taming Big Tobacco. Our first responsibility is always to our patients, but we also have a public responsibility to make chatbots safer for everyone.

Acknowledgments

None.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, Journal of Medical Artificial Intelligence. The article has undergone external peer review.

Peer Review File: Available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-0110/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://jmai.amegroups.com/article/view/10.21037/jmai-2026-0110/coif). A.F. serves as an unpaid editorial board member of Journal of Medical Artificial Intelligence from May 2026 to December 2027. The other author has no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Montero A, Montalvo III J, Kearney A. et al. One in Three Adults Have Used AI for Health Information and Advice in the Past Year. 2026. Available online: https://www.kff.org/public-opinion/kff-tracking-poll-on-health-information-and-trust-use-of-ai-for-health-information-and-advice/.
2. World Health Organization. World mental health today: Latest data. 2025. Available online: https://iris.who.int/server/api/core/bitstreams/31714489-1345-4439-8b37-6cbdc52e15ca/content.
3. Anderer S. About 1 in 5 US Youth Use AI Chatbots for Mental Health Advice. JAMA 2026;336:96. [Crossref] [PubMed]
4. Torous J, Blease C. Generative artificial intelligence in mental health care: potential benefits and current challenges. World Psychiatry 2024;23:1-2. [Crossref] [PubMed]
5. World Health Organization. Over a billion people living with mental health conditions—services require urgent scale-up. 2025. Available online: https://www.who.int/news/item/02-09-2025-over-a-billion-people-living-with-mental-health-conditions-services-require-urgent-scale-up.
6. Pinto da Costa M, Dragasek J, Iosue M, et al. Towards a more inclusive psychiatry: diversity among presenters and participants in the 25th World Congress of Psychiatry. World Psychiatry 2026;25:338-9. [Crossref] [PubMed]
7. Turner A. How Many Smartphones Are in the World? 2023. Available online: https://www.bankmycell.com/blog/how-many-phones-are-in-the-world.
8. Bielik J. ChatGPT still leads the AI model race, but Gemini and Claude are changing the scoreboard. Webiano Digital & Marketing Agency. 2026. Available online: https://webiano.digital/chatgpt-still-leads-the-ai-model-race-but-gemini-and-claude-are-changing-the-scoreboard/.
9. Frances A. Warning: AI chatbots will soon dominate psychotherapy. Br J Psychiatry 2025;228:474-8. [Crossref] [PubMed]
10. Brauer R, Alfageh B, Blais JE, et al. Psychotropic medicine consumption in 65 countries and regions, 2008-19: a longitudinal study. Lancet Psychiatry 2021;8:1071-82. [Crossref] [PubMed]
11. Strengthening ChatGPT’s responses in sensitive conversations. 2025. Available online: https://openai.com/index/strengthening-chatgpt-responses-in-sensitive-conversations/.
12. Frances A, Ramos L. Preliminary Report on Chatbot Iatrogenic Dangers. Psychiatric Times. 2025. Available online: https://www.psychiatrictimes.com/view/preliminary-report-on-chatbot-iatrogenic-dangers.
13. Keshavan M, Torous J, Yassin W. Do generative AI chatbots increase psychosis risk?. World Psychiatry 2026;25:150-1. [Crossref] [PubMed]
14. Ethics and governance of artificial intelligence for health: Guidance on large multi-modal models. 2025. Available online: https://www.who.int/publications/i/item/9789240084759.
15. WPA Position Statement on Global Digitalization in Mental Health and Care/GA23.10.18. 2023 Sep. Available online: https://www.wpanet.org/wp-content/uploads/2025/09/18.GA23.10.18-Digitalization-MHC-WPA-Position-Statement-30.5.23-14.6.pdf.
16. Burns W. Europe’s Regulatory Failure on Generative AI and Mental Health. 2025. Available online: https://www.techpolicy.press/europes-regulatory-failure-on-generative-ai-and-mental-health/.