Multicalibration requires calibration for all efficiently computable subgroups, not just the major protected groups. This addresses intersectional fairness by ensuring that even small intersectional subgroups receive calibrated predictions.