Basically, what we have here is a weakness in the client, namely a provision that allows the server to send the client a fresh key and ask for re-encryption and re-sending with the new key. This, in turn, would allow for a good old MITM attack if the server were to be compromised.
This re-encryption and re-sending of messages would be without intervention by the user, though a message "new key" would be displayed to the user provided they had chosen the option to display such notifications (which are disabled by default).
What's unclear to me is whether only messages that have not yet been delivered would be affected, or all.