SURE: A New Privacy and Utility Assessment Library for Synthetic Data

This paper introduces SURE, a comprehensive open-source library designed to assess the privacy risks and utility trade-offs of synthetic datasets. SURE addresses critical privacy concerns associated with synthetic data, such as the potential for individual identification through techniques like membership inference attacks. By offering a user-centric framework for evaluating both the statistical properties and privacy guarantees of synthetic data, SURE aims to balance the privacy-utility conundrum that often affects data anonymization efforts. The library provides robust tools for data scientists and compliance officers to ensure that synthetic datasets preserve both utility and privacy for effective AI training and data analysis, while adhering to GDPR and other regulatory standards. SURE's functionalities include statistical similarity tests, machine learning utility evaluations and privacy risk assessments, all of which are accessible through a user-friendly Python interface. Following extensive testing and validation, the finalized version of SURE will be available open-source.