The Adversarial Robustness Toolbox (ART) is a Python library for machine learning security. Developers and researchers use it to test and defend models against four kinds of threat: evasion, poisoning, extraction and inference. ART supports frameworks such as PyTorch, TensorFlow and scikit-learn and works with data types including images, tables and audio. It is a library you install yourself; there is no separate cloud service.
ART was originally developed by IBM and moved in 2020 to the LF AI & Data Foundation of the Linux Foundation. The code is available under the MIT licence. Because the library runs locally, models and test data stay on your own infrastructure and no vendor processes data. For European organisations this means little dependency on a cloud service, but a dependency on a project hosted by the Linux Foundation and originally contributed by IBM. The MIT licence allows you to maintain your own fork.