autoloader expects new files, not updates/overwrites of old files.

So basically autoloader will look for new filenames in the directory and process all new files.
And those files are (depending on settings/size) processed file by file or all at the same time.

If you want to make sure you have no dups, you will have to create a function that is processed with each batch and called by forEachBatch.
https://docs.databricks.com/en/structured-streaming/foreach.html