Abstract
Even when synthetic microdata contain no real records, an attacker may still infer sensitive attributes of real individuals if the synthetic data reproduce predictive relationships too closely — a form of attribute disclosure. This preprint introduces RAPID (Risk of Attribute Prediction-Induced Disclosure), a measure that quantifies exactly this risk for synthetic data. RAPID assesses how well sensitive attributes can be predicted from released variables, giving data holders a concrete, interpretable metric to set alongside utility when deciding whether synthetic data are safe to publish. The work reflects the SwissAnon group’s current research on rigorous risk evaluation for synthetic data.
synthetic data disclosure risk attribute disclosure privacy