Google's Differential Privacy Libaries

Google's Differential Privacy libraries are an open-source collection of tools for producing statistics from sensitive datasets while protecting the privacy of individuals. Differential privacy works by adding carefully calibrated random noise to results such as counts, sums and averages, so that no single person's data can be identified. Implementing it correctly is difficult, and subtle errors can weaken the guarantee. The libraries give organisations tested, ready-made components to apply differential privacy reliably, from research to production.

At the core are building block libraries in C++, Go and Java. They implement the standard Laplace and Gaussian mechanisms for adding noise, using secure noise generation. They also provide private versions of common statistics, including count, sum, mean, variance and quantiles. Two end-to-end frameworks build on these libraries and are designed to be usable by non-experts. Privacy on Beam serves Go users, and PipelineDP4j serves Java, Kotlin and Scala users. Both run on large-scale data processing systems such as Apache Beam and Apache Spark. They automatically limit how much each person can contribute to a result, which differential privacy requires.

Further tools support safe use. A privacy accounting library tracks the total "privacy budget" spent across multiple analyses. A stochastic tester helps detect code changes that could break the privacy guarantee, and DP Auditorium audits whether guarantees actually hold. A command-line tool runs differentially private SQL queries. The documentation openly lists known limitations. The libraries are released under the Apache 2.0 licence.

Auto-discovered on 2026-08-05 by OECD Catalogue Automation