UNESCO Recommendation on the Ethics of AI PA11a: paras 121 to 124 Health: regulate health AI to be safe, effective and proven, with bias oversight, clinicians and patients in development, consent, human final decisions and ethics review

Member States should use effective AI to improve health consistently with human rights, respect patients' relationships with family and staff, and regulate health and mental health AI (with attention to children and youth) to be safe, effective, efficient and scientifically and medically proven, involving patients in development. Prediction, detection and treatment AI should be regulated to ensure oversight of bias, inclusion of the professional, patient, caregiver or user as a domain expert, privacy and data protection, informed consent without blocking access to care, human final decisions on diagnosis and treatment, and ethics committee review before clinical use where necessary; research on mental health harms of AI should be established.