Abstract Gaia DR2 5932173855446728064 was recently proposed to be unbound from the Milky Way based on the $-614.3± 2.5\, \mathrm{km}\, \mathrm{s}^{-1}$ median radial velocity given in Gaia DR2. We obtained eight epochs of spectroscopic follow-up and find a very different median radial velocity of $-56.5 ± 5.3\, \mathrm{km}\, \mathrm{s}^{-1}$. If this difference were to be explained by binarity, then the unseen companion would be an intermediate-mass black hole; we therefore argue that the Gaia DR2 radial velocity must be in error. We find it likely that the spectra obtained by Gaia were dominated by the light from a star $4.3\, \mathrm{arcsec}$ away, and that, due to the slitless, time delay integration nature of Gaia spectroscopy, this angular offset corresponded to a spurious $620\, \mathrm{km}\, \mathrm{s}^{-1}$ shift in the calcium triplet of the second star. We argue that such unanticipated alignments between stars may account for 105 of the 202 stars with radial velocities faster than $500\, \mathrm{km}\, \mathrm{s}^{-1}$ in Gaia DR2 and propose a quality cut to exclude stars that are susceptible. We propose further cuts to remove stars where the colour photometry is suspect and stars where the radial velocity measurement is based on fewer than four transits, and thus produce an unprecedentedly clean selection of Gaia radial velocities for use in studies of Galactic dynamics.