Imagine a safety-net hospital in the Midwest being courted by vendors promising artificial intelligence (AI) tools to reduce readmissions and streamline prior authorization. On paper, it is precisely the kind of organization that should benefit from improved risk prediction and more efficient workflows: a high Medicaid share, chronic staffing shortages, and a patient population facing multiple social risks. In practice, however, the hospital has a handful of overextended analysts, unreliable data connections between its electronic health record and billing systems, and no in-house machine-learning expertise. The compliance team is already juggling new health-equity reporting requirements. Now it is expected to evaluate algorithmic bias, monitor model drift, and reassure their board that AI will not deepen disparities.