diffprivlib

IBM’s open-source Python library for training machine learning models and analyzing data with differential privacy guarantees.

Diffprivlib provides a high-level interface for data scientists to implement differential privacy using familiar tools like Scikit-learn. The library includes a suite of 15+ pre-built mechanisms (such as Laplace, Gaussian, and Staircase) and specialized versions of standard models like Logistic Regression and K-Means. By injecting calibrated noise into computations, it ensures that individual data points remain anonymous even under sophisticated re-identification attacks. It is designed to integrate directly into existing pipelines, requiring minimal code changes to transition from standard Scikit-learn estimators to privacy-preserving variants.

https://github.com/IBM/differential-privacy-library