Imagine a pharmaceutical company distributing an antidepressant through an app, letting users self-select their dose, with no licensed prescriber, no contraindication screening, and no adverse event reporting. The institutional consequences would be swift and severe as regulators have the authority and the obligation to act. Now remove the word “pharmaceutical” and replace it with “AI.” The rest of the description holds but the accountability does not. There are currently thousands of mental health apps and AI tools available for download, the overwhelming majority carrying no clinical validation that they work and no evidence that they are safe. The field has looked at this and decided it is acceptable. It is not. Mental health AI is distributing something that acts on psychological states at clinical scale in populations carrying diagnosable conditions, without the evidentiary requirements, deployment constraints, or accountability mechanisms that any other clinical intervention faces.