minutes

Resource

AI notetakers and attorney–client privilege

Privilege has one load-bearing wall: confidentiality. An AI notetaker that streams your client conversations to a vendor’s cloud puts a third party inside that wall, and nobody — not the vendor, not the bar, not yet a court — has given lawyers a clean answer on what that does to privilege. Here is the current state of the analysis, and the architectural choice that makes most of it moot.

Last reviewed: 2026-07-11Not legal advice

Quick answer

Waiver is fact-specific and unsettled — but the risk vector is not. Privilege requires confidentiality; a cloud notetaker is a disclosure to a third party under that vendor’s terms. Whether that disclosure defeats privilege will depend on the vendor’s data handling, your diligence, and a judge — three things you don’t fully control.

On-device transcription removes the third party entirely. A transcript generated and stored on the attorney’s own machine involves no outside disclosure to argue about. The confidentiality question collapses back to the one you already answer daily: is your laptop secure?

What The Bar Has Actually Said

The closest authoritative guidance is ABA Formal Opinion 512 (July 2024) on generative AI. Its core demand is unglamorous: before client information goes into an AI tool, the lawyer must actually understand the tool’s data handling — who can access inputs, whether they train models, how long they’re retained — and in some cases obtain informed client consent. Competence about the tool is part of the duty of confidentiality, not separate from it.

Applied to meeting notetakers, that means reading the vendor’s terms the way opposing counsel would: Does the vendor’s staff have access to transcripts for “service improvement”? Are recordings used for model training, by them or their subprocessors? What happens to your clients’ conversations under a subpoena served on the vendor? Every “yes” and “unclear” is material you may someday have to defend.

Separately from privilege, recording consent law applies: some states require all-party consent to record a conversation, others one-party. For client meetings, explicit consent is the professional norm regardless of what your state minimally permits — the RCFP guide linked below has the state-by-state map.

The Vendor Questions That Matter

If you’re evaluating any cloud notetaker for client work, get written answers to:

  • Where is audio transcribed, and by which subprocessors?
  • Is any human review of transcripts possible, ever, for any purpose?
  • Are recordings or transcripts used to train models — theirs or anyone’s?
  • What is the retention period, and is deletion verifiable?
  • What is their process when they receive legal process for your data?
  • Will they sign confidentiality terms that survive their standard ToS?

Notice what this list is: a due-diligence file you must build and maintain for a vendor whose entire involvement is optional.

The Architectural Answer

Every question above exists because the conversation leaves your machine. Run transcription on-device instead, and the third party disappears: no vendor receives the communication, no terms of service govern it, no subpoena served on a vendor can reach it, and there is no “service improvement” clause to parse. That is the design of Minutes — open source, on-device transcription and diarization, transcripts as markdown files on your own disk with owner-only permissions. The privilege analysis returns to familiar ground: secure the device, control access by matter, and document client consent to recording.

What on-device does not do: it doesn’t satisfy consent law for you, secure an unencrypted laptop, or make a transcript less discoverable in litigation — your own files are always discoverable. It removes the third-party disclosure question, which is the one you can’t fix after the fact.

Next step

Common questions

Does using an AI notetaker waive attorney–client privilege?
There is no blanket answer — waiver is fact-specific and courts have not squarely resolved cloud AI notetakers. The risk vector is clear, though: privilege depends on confidentiality, and sending client conversations to a third-party cloud service is a disclosure that opposing counsel can probe. Vendor terms allowing human review or model training make the argument harder to defend.
What does ABA Formal Opinion 512 say about AI tools and confidentiality?
ABA Formal Opinion 512 (July 2024) addresses generative AI under Model Rule 1.6: lawyers must evaluate a tool's data handling before inputting client information, may need informed client consent for self-learning tools, and remain responsible for understanding where client data goes. It does not ban AI tools; it requires lawyers to actually understand the data flow.
Do I need consent to record client meetings?
Recording-consent law is separate from privilege and varies by state: some states require all-party consent, others one-party. Best practice for client conversations is explicit consent regardless of state law — it is also the professional norm.
How does on-device transcription change the privilege analysis?
If transcription runs entirely on the attorney's own machine and the transcript is a local file, no third party ever receives the communication — there is no vendor disclosure to analyze, no terms of service governing client data, and no vendor to subpoena. The remaining duties are the familiar ones: device security and access control.

Sources

This page is informational, not legal advice. Privilege and consent questions are jurisdiction- and fact-specific — consult your ethics counsel before adopting any recording tool for client work.

Keep reading