We study inflation in a random multifield potential, using techniques developed by Marsh et al. The potential is a function of a large number of fields, and we choose parameters so that inflation only occurs in regions where the potential is accidentally flat. Using an improved estimate for the dynamics of eigenvalue repulsion, we are able to describe the steepening of the potential as inflation progresses. We provide suggestive arguments, but not a proof, that the resulting scalar power spectrum generically disagrees with observations. We also point out two problematic aspects of the model: there is no well-defined probability distribution for the gradient of the potential, and the evolution of the potential over small distances in field space is unphysical.