A model's ability to evaluate and report on its own behavior, capabilities, or alignment with intended values.