5/22/26 Microsoft is pushing AI safety closer to a true engineering discipline with the open sourcing of RAMPART and Clarity. RAMPART brings automated red-team testing for agentic AI into CI/CD pipelines, helping organizations continuously test for prompt injection, unsafe tool use, and behavioral drift at scale — including multi-turn attack scenarios and probabilistic safety thresholds.
Clarity tackles a different challenge: forcing teams to think through architecture, risks, and unintended consequences before code is written. Together, the tools highlight a growing reality for security leaders — securing AI agents requires both continuous adversarial testing and disciplined design governance from the start.
https://www.theregister.com/security/2026/05/21/microsoft-open-sources-agentic-ai-safety-tools/5243822