The file is scanned by ClamAV before anything else opens it. If it fails the scan it is rejected and never decoded.
It is transcribed on this server by whisper.cpp running locally. Nothing is sent to any third party, and the machine has no outbound network access from the service that does the work.
The audio and the transcript are deleted as soon as the response is sent. Neither is written to a log — only the duration and how long it took.
Accuracy varies with recording quality and accent. Treat it as a first draft, not a record.