Last released Oct 1, 2026
Measure and guard AI agents: safety and quality evaluation with a runtime guardrail