Microsoft has open-sourced two tools, Rampart and Clarity, designed to integrate AI safety checks into the agent development lifecycle. Rampart, built on PyRIT, enables continuous automated red-team testing within CI/CD pipelines, targeting issues like cross-prompt injection, unsafe tool execution, and privilege escalation. Clarity operates earlier in the workflow, helping engineers validate design assumptions, define trust boundaries, and document decisions as markdown files in the repository. Both tools are part of Microsoft's broader agent governance strategy, which also includes the recently released Agent Governance Toolkit targeting OWASP risks for AI agents.
Sort: