Microsoft Releases Open Source AI Safety Tools RAMPART and Clarity

Microsoft announced in a blog post the release of two open source tools, RAMPART and Clarity, aimed at improving AI agent safety during development. The company said the tools bring security and safety checks into the engineering workflow, allowing developers to identify and address risks earlier in the lifecycle.

RAMPART is a testing framework that enables teams to encode adversarial and benign scenarios as repeatable tests. Built on PyRIT, Microsoft's automation framework for red teaming generative AI systems, it allows engineers to run safety tests in continuous integration environments. The framework supports statistical trials to reflect the probabilistic behavior of large language models and helps teams preserve lessons from red team exercises by converting findings into reusable tests.

Clarity focuses on the design stage of AI agent development. It guides engineers through structured discussions on problem definition, solution exploration, failure analysis, and decision tracking. The tool records the results as markdown files within a project repository, allowing them to be reviewed and updated like source code. Clarity also uses multiple AI evaluators to analyze a system from perspectives such as security, human factors, and operational reliability.

Both tools are available now as open source projects. Microsoft said they are part of a broader effort to integrate AI safety practices directly into the engineering process, ensuring that safety considerations evolve alongside system development.