I'm also working with AutoLoader these days to create an ingestion pattern and troubleshooting it can be tricky.
I wonder if you could pick a single file (whose full path / location / uri you exactly know) and read it without autoloader. Just with spark.read.json ... or SELECT * FROM JSON.`path`
If that fails then it's not an autoloader issue. If that works, that might help further troubleshooting. Sometimes the order of the .option blocks can have an impact (it shouldn't ...), I experienced that just yesterday so keep an eye out for that, too.