Consider protocol like QUIC or MinimaLT which are essentially layer4 replacement for TCP/UDP (riding on top of UDP currently because of fear of random firewalls dropping new L4 protocol).
From crypto point of view they are mostly the same designs (using same EC stuff djb has done). IP addresses are not relevant for identity anymore as identity is established via use of PKI.
This is great, long-lived session can move from one IP to another one, and session can be kept up, as IP address does not matter.
However I don't think either of them expose identity to MITM. I think there might be quite nice use-case in your own home FW or corporate FW to say 'Alice can connect to server X port 22'. Then regardless what IP Alice is using, as long as she offers the specified identity, she is passed by the FW.
Could it be possible for arbitrary MITM to securely inspect identity so that no other arbitrary MITM could have faked it?