We definitely do not have any centralized servers. The currently-specified mechanism to determine routes is that they are decided by connecting to a distributed peer-to-peer network that is modeled in many ways after a distributed hash table, at which point you do lookups of random keys in the hash space, which results in a randomized scalable lookup. There are mechanisms in play (and which I assume are already described in our in-progress whitepaper) to protect against eclipse attacks, sybil attacks, and some other miscellaneous attacks on the routing infrastructure. That said, this is also some ongoing work, and we are actively engaging with and being reviewed by people both in academia and in the field with respect to our techniques.