A new study reveals limitations in the current use of mathematical models for personalized medicine, particularly in schizophrenia treatment. Although these models can predict patient outcomes in specific clinical trials, they fail when applied to different trials, challenging the reliability of AI-driven algorithms in diverse settings.