This project explores how the specific nature of regulations and guidelines for AI in healthcare as decision-support systems is generating an experimental space where middle managers and expert users (physicians, radiologists) must make sense of the AI’s biases, over-alerting, and lack of uniform standards for monitoring and evaluation. We explore in this context the emergence of bottom-up usage practices, highlighting the pivotal role of middle managers and radiologists in redefining boundaries during an innovative, early adoption process.