SAFER-MH: Improving Safety Evaluation and Regulatory Capacity for AI-Enabled Mental Health Care

South Africa’s medicines regulator is getting a new framework to evaluate AI-powered mental health apps and devices, alongside tools to detect the specific harms these technologies can cause. Mental health apps and chatbots are proliferating, but regulators have no established way to assess their safety or effectiveness. The same AI that can triage depression symptoms or deliver therapy could also give dangerous advice, reinforce biases, or mishandle sensitive user data. Without clear rules, developers either avoid the market or launch products with unknown risks. This project tackles that gap head-on. The researchers will build a regulatory pathway with SAHPRA, define a taxonomy of AI-specific harms in mental health, and create a standard for logging user engagement data so outcomes can be tracked. The work focuses on low-resource settings, where AI could expand access to care but regulatory capacity is thinnest. If successful, the framework could reduce the cost and complexity of compliance for developers while giving regulators a repeatable method to evaluate safety. Because the field is so new, the outputs—harm taxonomies, data standards, regulatory templates—will likely be adopted globally, shaping how AI is governed in mental health from the ground up.