Secure aggregation in federated learning lacks strong privacy against membership inference attacks, necessitating additional privacy mechanisms.