I was under the impression that attack would not work because as long as the majority of nodes are "honest" there would be no way for an attacking node to outpace the chain. Each node votes with their cpu so the combined compute/speed of honest nodes would always outpace the introduction of a corrupt block. Since the the most recent "confirmed block" is a hash of the previous successful block it just continues.
However I admit much of my assumptions are just from reading the original paper so i'm probably over-simplifying. I guess it would make sense they need some history but not sure why they would need all of it since the previous blocks are confirmed?
