How do they detect if someone poisons their data, if they not at least associate IP addresses to the data?