We roughly know the space of possible mutations in the human genomes because we have so many sequenced genomes now: if we don't spot a mutation it's probably not good when it happens, survivorship bias.

The problem with CRISPR is that we cannot control where the off-target effects happen, we can currently only optimise the guiding RNA and the Cas enzyme to have as little off-target effects as possible (but not 0, yet). It would be cool to engineer guiding RNAs that bind in those high mutation-rate areas when they have off-target effects, stuff can mutate there and nothing will happen (probably).