Arize Model Drift is a machine learning observability platform that monitors and detects drift in production AI models. The platform tracks changes in model inputs, outputs, and actuals over time to identify when models have grown stale or are experiencing performance degradation. The platform monitors four main types of drift: prediction drift (changes in model predictions), concept drift (changes in actuals), data drift (changes in feature distributions), and upstream drift (changes in data pipelines). It uses statistical distance measures to quantify distribution changes between baseline and current data. The platform provides monitoring capabilities for models with delayed or no ground truth, allowing teams to detect issues before they impact business outcomes. It includes explainability features using SHAP values to understand drift causes and supports monitoring of unstructured data. Arize Model Drift enables teams to diagnose production issues, identify data quality problems, detect adversarial inputs, and understand when model retraining or replacement is needed. The platform compares distributions between training, validation, and production data to measure drift across different time periods.