A type of cybersecurity threat where an attacker uses access to a machine learning model’s predictions to infer sensitive information about the model’s training data. This can potentially lead to privacy breaches, revealing personal or proprietary data that was used to train the model.