Not yet reviewed by Pith; the record is open.

We update a flux-limited complete sample of Swift-based SGRBs (SBAT4, D'Avanzo et al. 2014), bringing it to 25 events and doubling its previous redshift range. We then evaluate the column densities of the events in the updated sample, in order to compare them with the NH distribution of LGRBs, using the sample BAT6ext (Arcodia et al. 2016). We rely on Monte Carlo simulations of the two populations and compare the computed NH distributions with a two sample Kolmogorov Smirnov (K-S) test. We then study how the K-S probability varies with respect to the redshift range we consider. We find that the K-S probability keeps decreasing as redshift increases until at z$\sim$1.8 the probability that short and long GRBs come from the same parent distribution drops below 1$\%$. This testifies for an observational difference among the two populations. This difference may be due to the presence of highly absorbed LGRBs above z$\sim$1.3, which have not been observed in the SGRB sample yet, although this may be due to our inability to detect them, or to the relatively small sample size.