Attribute Inference Attack

Definition: An attribute inference attack aims to infer unknown or sensitive attributes of individuals in a dataset using the outputs of a machine learning model and partial knowledge of their other attributes.

Context/Example: Even if a model does not explicitly use “medical condition” as an input, an attacker who knows a user’s age, location, and habits may use the model’s predictions to infer whether the user likely has a certain condition.

Related Concepts: Privacy Leakage, Feature Correlation, Differential Privacy