Can AI be more careful about what you say?