Evaluatingedge cases—rare but critical scenarios where AI may behave unpredictably—must be done during thetest and verifystage (D). This phase is designed to simulate extreme or unusual inputs, validate performance under stress, and ensure robustness and safety before deployment. AAIA highlights thatrobustness testing, including edge case evaluation, is a key testing technique for AI solutions.

While planning (A) and data collection (C) prepare inputs, they do not evaluate model behavior. Operating and monitoring (B) may detect issues later, but edge case testing must occurpre-deploymentto reduce risk.