Cross-Silo De-Anonymization Under Local Differential Privacy: Threat Model, Phase Transition, and Coordination Necessity
A new research paper introduces a framework called cross-silo person-level DP (XSP-DP) to analyze de-anonymization risks when data is split across multiple silos, each protected by differential privacy. The study identifies a critical threshold, approximately Theta(log n / epsilon^2), where de-anonymization becomes feasible as the number of data silos increases. The findings suggest that even if individual data silos offer privacy, combining information from multiple silos can lead to inevitable de-anonymization without explicit coordination. AI
IMPACT Establishes a theoretical threshold for de-anonymization risks in distributed private datasets, impacting future privacy-preserving system designs.