The vendor has to hear it
Cloud transcription cannot be end-to-end encrypted. The server must process the audio in the clear, so the vendor and its subprocessors hold readable recordings.
MDN Web Speech API · Zoom E2EE documentation
Why local
Cloud dictation hands your client's words to someone else.
Most dictation tools, including the ones built into phones and Windows, send audio to a server to be transcribed. Encryption protects it on the way there. It does not stop the vendor from reading it once it arrives.
Defaults favor keeping it
Several major services retain voice data or use it to improve their models unless you opt out. Human review of recordings by contractors has been documented more than once.
AWS Transcribe docs · Apple, Microsoft privacy pagesCourts are paying attention
A 2026 federal ruling treated a client's use of a consumer AI tool as disclosure to a third party. Bar opinions now ask lawyers to know where recordings are stored and who can use them.
U.S. v. Heppner (S.D.N.Y. 2026) · NYC Bar Formal Op. 2025-6
Side by side
Where your audio goes
| Question | Typical cloud dictation | IdeaStream server |
|---|---|---|
| Where is speech turned into text? | The vendor's data centre | A machine in your office |
| Who can access the audio? | The vendor, its subprocessors, and anyone with a court order against them | Your firm |
| Is it used to train models? | Depends on the plan and your settings | No |
| Can you prove it was deleted? | Usually a request and a 30-day wait, rarely a certificate | You delete the file yourself |
| Does it work without internet? | No | Yes |
Find out where your firm's audio goes.
A 20-minute call. We'll ask what your attorneys use to dictate and tell you plainly whether an audit is worth it.
Book a call
[you@yourdomain.com]