AI decision-making systems show promise but lack consistent evidence for improving patient-relevant outcomes like mortality and quality of life. Rigorous harm-benefit assessments and transparency are needed for safe integration into healthcare.