I think the idea is to avoid NHEJ entirely. I would guess that it's going for base-excision-repair (BER), which is why it's currently targeting either C or G bases.

If the wrong guide sequence is bound you're screwed no matter what, though I suspect they'd say that the error rate is low enough to be practically irrelevant. Not sure that's really true though, the genome is pretty big and if you screw up at even astonishingly low rates, you're still screwing up fairly frequently.