Microsoft ASSERT Tool Automates AI Behavior Testing

Microsoft has launched ASSERT (Adaptive Spec-driven Scoring for Evaluation and Regression Testing), an open source framework that converts plain-language descriptions of AI behavior goals and policies into structured, scored test cases. It also tracks the paths AI systems take during testing, helping developers pinpoint failures. The tool addresses a gap in existing evaluations, which tend to be broad rather than application-specific. Developers can define custom rules and constraints, and ASSERT generates ongoing tests to verify compliance. Microsoft says it can be used during development, after deployment, and for continuous monitoring.