The ability of an AI system to change its behaviour or its relevant evidence after an evaluation. This may arise from model updates, configuration adjustments, new data or drift in the user population.