Privacy Preserving and Verifiable Data Access Using Differential Privacy, Authenticated Encryption, and Cryptographic Integrity Verification

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Sensitive information systems must protect stored records from unauthorized disclosure and modification while still permitting legitimate statistical analysis. Encryption alone does not control inferential disclosure arising from authorized aggregate queries, and differential privacy alone does not protect raw records at rest. This paper presents an implementation oriented framework that keeps these responsibilities separate while integrating them into one system, using AES 256 GCM authenticated encryption, a SHA 256 Merkle tree with RSA PSS root signing for dataset level integrity, role based access control, a custom Laplace differential privacy mechanism, and a persistent PostgreSQL backed privacy accountant. The prototype is built in Java 21, Spring Boot 3.x, and PostgreSQL 16.4, and is evaluated on synthetic Synthea generated healthcare records so that no real patient information is processed at any stage. Across seven controlled tampering scenarios and 700 total trials, every attempt was detected, with detection attributed separately to AES GCM authentication, Merkle root verification, and RSA PSS signature verification. Differential privacy evaluation across five query classes at epsilon values from 0.1 to 8 showed mean absolute error and root mean square error decreasing monotonically as epsilon increased, and the persistent privacy accountant correctly enforced sequential composition, rejected over budget requests, and retained its state across an application restart and under concurrent access. On the 10,000 record dataset, the complete framework recorded a mean end to end latency of 46.72 milliseconds compared with 24.91 milliseconds for an unprotected baseline, a statistically significant but practically bounded overhead. The contribution of this work is the integrated architecture, implementation, and empirical evaluation of these mechanisms within one framework, not the invention of the underlying primitives, each of which is adopted from established, cited sources. An earlier Triple DES and RSA implementation is retained only as legacy context, consistent with NIST's withdrawal of the Triple DES recommendation in 2024 [6].