Researchers found that medical AI models are susceptible to membership inference attacks (MIAs). Attackers can identify if a patient's data was used in training, potentially exposing sensitive medical history. The study calls for stricter privacy audit standards and the use of differential privacy.