I guess they don't really know what they're doing and are learning how load balancing works on the fly, and thinking that'll result in HA without side-effects without further work.

EDIT: Don't really get why this was downvoted. With the proper technical knowledge it's clear to anybody that two instances with different JWT secrets behind a load balancer is going to cause this very issue. So the fact that they set it up that way means they have a knowledge gap ("they don't really know what they're doing"). At the very least they should enable sticky sessions on the load balancer if they insist on going this route, which would mitigate the issue (but depending on client-side configuration would not necessarily prevent it completely).

Don't take this as an insult towards the admins of the instance, I'm just pointing out there's a lack of knowledge, and some trial-and-error going on.