Why sensitive recordings should never leave your Mac

On-device transcription runs the speech model on your own hardware. No audio file is packaged up and sent to an API, no server-side queue processes it, no vendor stores a copy while it works. The file goes in, the transcript comes out, on the same machine.

That is a different guarantee than "we delete your data after 30 days" or "encrypted in transit." Those are policies - promises about what a third party will do with something they already have. On-device transcription never gives them anything to have.

Some audio is fine on a cloud server. A public podcast episode, a YouTube video - already broadcast, no confidentiality to lose. A client meeting, a therapy session, a legal deposition, a medical consult is a different category: information a third party has no business holding, however good their security page reads.

The test is simple. If you would not paste the content into a public forum, think twice before uploading the audio to a company whose retention policy you have not read.

On-device transcription works with no network connection at all - on a plane, in a basement server room, anywhere. There is no API call to fail, no rate limit, no outage on someone else's status page holding up your work.

It is also the more defensible posture under GDPR and similar data-protection rules: audio and transcript never cross into a processor's infrastructure, so there is no data-processing agreement to negotiate and no additional party in your compliance picture for this step.

The transcript lands in a folder on your disk, not behind a login on someone else's platform. You export it, move it, delete it, back it up - the same as any other file you own, because it is one.

Cloud transcription is usually billed per minute, which turns a large archive into a running cost. On-device has none of that: transcribe one file or ten thousand, on your Mac, and the price is the same - the compute you already paid for.