Testing techniques to evaluate the behavior of machine learning models and AI systems when receiving inputs outside the training data distribution. Detects anomalies, drift, edge cases, and vulnerabilities related to unforeseen data, essential for model robustness and security in production.