Microsoft ASSERT: Test AI Agents with Plain Text Specs
Summary
Microsoft's ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing) is an open-source, local-first framework released at Build 2026. It enables developers to transform natural language policy requirements into structured, executable test cases for AI agents. By automating taxonomy generation, stratified testing, and trace-grounded scoring, ASSERT provides verifiable metrics and policy citations for agent behavior. It is framework-agnostic, supporting tools like LangChain and CrewAI, and is designed to work in tandem with the Agent Control Specification (ACS) to create a comprehensive, continuous lifecycle for AI governance and safety.
(Source:GlobalGPT)