We've had various issues all very similar in nature, never managed to get a root cause. Methods of fixing include -

1. Resume replication if it's paused
2. Remove database from secondary...

We had a similar issue but it only dropped the AG from one of the replica's recently which was bad enough.  Take a read of the following article and see...

Can I ask why you needed to rebuild the AG's?   We had issues where SQL dropped a Primary replica from the AG, the only solution was to drop and recreate...

We had a very similar issue where the Primary replica was dropped leaving our systems offline.  The following article explains the situations that can cause the replica to be removed...

I tried all the above and I still could not remove the tempdb file.  I was trying to fix the issue on an AlwaysOn secondary node and found that I...

Online index rebuilds could cause the MDF file to grow quickly and leave free space behind once it's completed

If I understand you right this should be possible. We have a readable secondary reporting instance with two AGs where the primary replicas for these AGs are both different servers.

Thanks, I came to the same conclusion that it must appear as IO on the hidden snapshot. Shame as I normally find the virtual file stats extremely useful.

In the end...