Microsoft Open-Sources RAMPART and Clarity — Safety Testing Tools for AI Agent Development Workflows

Microsoft Security Blog·medium signal

Microsoft Security published RAMPART (risk assessment for multi-agent protocols and runtime threats) and Clarity as open-source tools on May 20, designed to bring safety testing directly into agent development workflows. The tools address the gap between model-level safety evaluations and production agent behavior, where tool-calling agents can exhibit emergent unsafe behaviors not visible in isolated model testing. For teams deploying agents with real-world tool access, these provide structured threat modeling and runtime monitoring without building custom safety infrastructure.