PyRIT (Python Risk Identification Tool) is an open-source framework for automated and human-led AI red teaming, which security professionals and engineers use to proactively identify risks in generative AI systems. It supports targets such as OpenAI, Azure, Anthropic, Google, Hugging Face and custom HTTP endpoints, runs single-turn and multi-turn attacks (including Crescendo and TAP), scores responses with scorers and stores results in SQLite or Azure SQL. You install it with pip and use it through a scanner, a web interface or the Python framework.
PyRIT is MIT-licensed and maintained by Microsoft; the repository sits under the Microsoft organisation on GitHub. It is not a community-governed project. It runs in your own environment, so there is no vendor-hosted service that receives your data; the target models you test may, however, be hosted elsewhere. For a European organisation this means control over execution and test data, but dependence on Microsoft for the roadmap.