Privacy evaluation should not by default carry over conclusions from high-dimensional datasets, but needs to be recalibrated for the characteristics of tabular data.