Text this: Embedding Fear in Medical AI: A Risk-Averse Framework for Safety and Ethics