Secure network coding addresses wiretapping of an unknown subset of links chosen from a given collection of possible sets. This talk overviews some of our recent results generalizing beyond the well-known uniform case where any $z$ noiseless links can be wiretapped, and all randomness originates at a single source. We show that when other nodes can generate randomness, even a single unicast uniform wiretap problem is as hard as the non-multicast network coding problem. We also discuss separation between wiretap channel coding and secure network coding on networks of wiretap channels, and show that the capacity region under weak and strong secrecy requirements is the same.