IPM Take
This is one of the clearest examples of technology adoption outrunning health-system governance. People are already turning to general AI tools for emotional support, but many of these tools were not designed, tested or monitored for mental health use. The policy question is no longer whether AI can support mental health. It is whether governments, health systems and technology companies can define safety, crisis referral, accountability and cultural adaptation before vulnerable users are harmed.
Executive Summary
On 20 March 2026, WHO reported on an expert workshop held by the Delft Digital Ethics Centre at TU Delft, the first WHO Collaborating Centre on AI for health governance, including ethics. More than 30 experts in AI, mental health, ethics and public policy examined the rapid use of generative AI for mental health and well-being. The recommendations included recognising generative AI as a public mental health concern, integrating mental health into AI impact assessments and monitoring, grounding tools in evidence, and co-designing AI systems with mental health experts and people with lived experience, including youth.
Why it matters
- Regulators: Need to clarify when AI tools used for emotional support cross into health-related use and require safety, monitoring and accountability standards.
- Data / AI leaders: Should build mental health safeguards into product design, including crisis escalation, age-sensitive protections, evidence review and transparent limitations.
- Clinicians: Need to prepare for patients who are already using AI tools informally for distress, loneliness or advice, sometimes before seeking professional help.
Until recently, digital mental health policy focused mainly on regulated apps, telehealth and online therapy platforms. The situation has changed because people now use general-purpose AI chatbots for emotional support, distress, loneliness and advice, often outside clinical settings.
That creates a new governance problem. These tools may be available instantly, but they may lack clinical validation, crisis escalation pathways, age-sensitive safeguards, privacy protections and cultural or linguistic adaptation. The relevant population is broad, but the concern is strongest for young people and users in moments of emotional vulnerability.
The WHO-reported expert workshop does not present generative AI as automatically harmful or automatically beneficial. Its message is sharper: AI tools used for mental health support need governance before they become informal care infrastructure. That means impact assessment, monitoring, evidence standards, public-health oversight and co-design with mental health professionals and people with lived experience.
For IPM, the implementation message is direct. AI mental health support cannot be left to platform growth alone. If these tools are becoming part of the mental health pathway, they need clear rules on safety, accountability and referral before adoption becomes impossible to control.
