Microsoft has released ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing), an open source framework that lets developers create AI behavior tests from plain-language descriptions of their system's intended goals, policies, and constraints. The tool converts those descriptions into structured tests, generates and runs problem scenarios against the target system, scores the results, and records the AI's decision paths so developers can pinpoint failures. Microsoft says the framework can be used during development, after deployment, and for continuous monitoring, addressing the need for application-specific evaluations that broader benchmarks cannot cover.