I believe audio is normally processed on device but transcripts can be sent to apple.

Even apps like superwhisper I'm looking to ditch because during an update they swapped my local models to cloud models which is a crazy breach of trust.

Very happy to see this stuff!