The two datacenters 'A' and 'B' that has site-to-site VPN set up, has internal subnets that are colliding. To work-around that I saw that the network team has NATed the IP addresses at 'B' to public IP addresses. Everything works. So, instead of using private IP addresses, clients from 'A' connects to a public IP address in 'B', which in turns gets de-NATed to the original private IP at 'B'. So, when a user at 'A' wants to SSH to a server in 'B', that user SSHs to a public IP address. Network team justifies it by saying it is completely secure since the traffic flies over VPN even though the destination address is a public IP. Is it something widely done across the industry to work around colliding subnets? How good is that from security point-of-view.
Yes, it's done quite often. Using public addresses eliminates a lot of problems when connecting multiple networks which use the same addresses in their local networks. If you're connecting multiple networks with the same IP ranges on the inside, things can get very, very ugly with very complicated NAT setups.
Of course, only public IP addresses actually assigned to the networks should be used or you'll probably be blocking some part of the public internet.