There are errors associated with your measurements. When you get down to a low enough concentration, and you have calibrated correctly, then you can, statistically, get negative numbers. This is quite normal. As long as you are not getting large negative numbers then the data is telling you that the element(s) you are measuring are not present at your detection limit. This is also why you determine, and use, a lower limit of detection (LOD) and a lower limit of quantification (LOQ). These indicate the point at which your calibration is no longer valid; you shouldn't report values below your LOD.
there is a chance for the unknowns(low concentrated or having no analyte in the sample) to give negative results if the blank is different or use different volume/weight of blank to make up the samples.