Cluster lensing has become an important tool in the search for high-redshift galaxies through its ability to magnify sources. In order to determine the intrinsic properties of these galaxies, lensing mass models must be constructed to determine the magnification of the images. These models are traditionally 2D, focusing on the mass within the cluster and either ignoring or approximating any contribution from line-of-sight galaxies. In this paper, we present the first full set of 3D mass models of the six Hubble Frontier Fields and use them to test for systematic biases in magnifications due to using the traditional 2D approach. We find that omitting foreground or background galaxies causes image position offsets between 0.1 and 0.4arcsec, a non-negligible fraction of the typical 0.3-0.7arcsec residuals of current state-of-the-art models. We also find that median image magnifications can shift by up to 6 per cent, though it is dependent on the field. This can be alleviated in some cases by approximating the mass in the lensing plane, but a 5 per cent magnification bias still exists in other cases; image position offsets are also improved, but are still present at 0.10arcsec.