Differentially private consent storage and enforcement

Abstract

The increasing adoption of data protection regulations, such as GDPR, CCPA, and LGPD, has made the storage and enforcement of user consent a critical requirement for database systems. Consent metadata, however, introduces new challenges: it must not only enable precise enforcement of user choices but also remain private, since the opt-in/opt-out status itself can leak sensitive information. This thesis addresses this dual challenge by proposing novel data structures that both represent and protect consent metadata under rigorous privacy guarantees. First, we introduce Purpose Filter, a filter-based data structure that encodes purpose metadata with a zero false positive rate, ensuring that unauthorized access requests are never granted. While Purpose Filter improves storage efficiency and query integration, its probabilistic nature is insufficient to protect against set-reconstruction attacks. To overcome this, we develop the Differentially Private Counting Bloom Filter (DPCBF), a new construction that perturbs counters with integer-valued geometric noise, providing formal differential privacy guarantees while maintaining practical utility. Building on these foundations, we design the Differentially Private Purpose Filter (DPPF), which combines consent enforcement with differential privacy without violating consent constraints. We conduct extensive experimental evaluations, including analyses of filter behavior, comparisons with existing DP-Bloom Filters, resistance to reconstruction attacks, and applications in machine learning with consented datasets. Results show that our structures effectively balance privacy, utility, and compliance, offering a practical pathway for privacy-preserving consent enforcement in modern database systems.