Systematic errors in model outputs arising from data, objectives, or deployment context, affecting fairness.