On March 20, 2026, the World Health Organization published an expert consensus statement, “Towards responsible AI for mental health and well-being,” aimed at a phenomenon that is already happening with almost no governance: large numbers of people — young people especially — are using generative AI tools for emotional support, even though those tools were never designed for mental health and have never been clinically tested for it. The experts’ conclusion is blunt: this is no longer a product design question. It is a public health issue.
The consensus grew out of an online workshop on January 29, 2026, hosted by the Delft Digital Ethics Centre (DDEC) at TU Delft, bringing together more than 30 international experts across AI, mental health, ethics, and public policy. DDEC is the first WHO Collaborating Centre focused on AI for health governance, including ethics, and the workshop was an official pre-summit event of the India AI Impact Summit 2026.
A 30-Person Workshop, and a Standing Consortium
The document itself is only part of the story; the institutional scaffolding behind it matters more. In the same announcement, WHO said it is establishing a Consortium of Collaborating Centres on AI for Health, with member institutions spanning all six WHO regions. Candidate members had already met for a pre-convening at TU Delft on March 17–19, 2026. In other words, this statement is not a one-off declaration but the first output of a standing governance network — expect standards, evaluation frameworks, and cross-border coordination to follow.
Dr. Alain Labrique, WHO’s Director of Data, Digital Health, Analytics and AI, put it most directly: as AI increasingly interacts with people “in moments of emotional vulnerability,” these systems must put safety, accountability, and well-being first. Sameer Pujari, WHO’s AI lead, went further: “We are at a critical juncture,” and the pace of AI adoption in daily life “has far outstripped investment in understanding its impact on mental health.”
Three Core Recommendations
The consensus lays out three principal recommendations, each pointing at a concrete governance action:
- First, treat generative AI use as a public mental health concern, with coordinated responses from government, health systems, and industry. The scope deliberately covers all generative AI solutions — not only products marketed for mental health.
- Second, integrate mental health into impact assessments and ongoing monitoring of AI solutions, spanning health determinants, short-term clinical measures, and long-term outcomes.
- Third, co-design AI mental health tools with mental health experts and people with lived experience, including youth, grounding them in evidence and tailoring them to cultural, linguistic, and contextual factors, while empowering consumers.
Why Frame It as a Public Health Issue
The load-bearing phrase is “all generative AI solutions.” Existing regulatory frameworks mostly treat medical-purpose AI as high-risk and regulate it tightly — the EU AI Act is the canonical example. But the risk source the experts name is the general-purpose chatbot: it sits outside medical device rules, yet in practice absorbs an emotional support role. Dr. Caroline Figueroa of TU Delft, a workshop participant, called for consensus on crisis referral frameworks and accountability systems; the workshop summary also flags long-term risks, including emotional dependence on AI companionship.
That is exactly why the public health framing matters. When risky behavior happens inside “non-medical” products, the traditional medical-device regulatory path never sees it. A public health lens looks at exposure, consequences, and intervention points at the population level instead. The speed at which AI has embedded itself in daily life — the running theme of the industry’s 2026 opening — is now clearly outrunning governance, and this document is WHO’s formal response to that gap.
What It Means for Product Teams
Three direct effects. First, if you build general-purpose AI products, stop assuming “we’re not a medical product” means “mental health risk is not our problem.” WHO’s framing effectively pulls emotional-companionship use cases into the scope of product responsibility. Second, impact assessments need mental health indicators, and crisis detection plus human referral paths will shift from nice-to-have to baseline — particularly for products used by teenagers. Third, co-design is not PR language here: the document explicitly requires involvement of mental health experts and people with lived experience, which raises the minimum bar for research and design processes.
For developers and product owners, the practical move is to treat this consensus as a checklist for the next two years of regulatory and reputational risk. How your product behaves in someone’s most vulnerable moments is a question you will eventually be asked — by a regulator, a journalist, or both.
Sources
- Towards responsible AI for mental health and well-being — WHO
- WHO issues three recommendations for responsible AI use in mental health — HTN
AI-assisted summary compiled from the sources above, reviewed by a human before publishing.