A bank runs any-to-any voice between branches over private MPLS, must encrypt without new tunnel headers or routing changes, and needs scalable rekey to all encryptors at once. Which site-to-site VPN group key design fits the private transport constraint?

The correct design succeeds because key servers authenticate group members and distribute one shared IPsec security association with rekey policy over private multicast, so every branch encryptor uses identical keys while preserving the original IP header, leaving MPLS forwarding and existing routing untouched and enabling one-to-many rekey ideal for low-latency any-to-any voice without tunnel overhead. The DMVPN design fails on forwarding concept because GRE encapsulation plus IPsec adds new outer headers and requires NHRP resolution with routing changes for dynamic spoke-to-spoke tunnels, violating the native-header constraint even though it runs over private MPLS. The FlexVPN design fails on key-management concept because IKEv2 negotiates pairwise point-to-point security associations that must be built and rekeyed per peer, so it cannot rekey all encryptors at once with a single group operation. The VTI design fails on scaling and routing concept because each point-to-point virtual interface needs tunnel state plus BGP or static routing to direct traffic, creating per-tunnel management and pairwise rekey instead of true group encryption. As an exam caveat, GET VPN requires end-to-end private IP reachability and multicast or cooperative rekey for key delivery and is unsuitable for Internet-only transport without those services. As a practical check, confirm key-server reachability, matching group identity ACL and transform set on all members, and that voice DSCP and multicast are preserved across MPLS.