Use Cases
AI Meeting Notes and Transcription for Lawyers: Consent, Privilege and Setup
AI meeting notes are ethical for lawyers only with all-party consent, a no-training data path and a retention rule. Bar opinions, tools and a file-note prompt.
On this page
- The rule in one line: off by default, on by consent
- What the bar opinions actually say
- Which tool, which data path
- Where privilege is at risk
- Consent in Europe: data protection and criminal law
- The setup that works
- The prompt: transcript into file note and action list
- Retention: decide before you press record
- Where this leaves you
Lawyers may use AI to record, transcribe and summarise meetings, but only with the informed consent of every participant, a data path that excludes model training, and a retention rule written before the first recording. The New York City Bar's default, set out in Formal Opinions 2025-6 and 2026-2, is not to record at all; everything below is about earning the exception.
The rule in one line: off by default, on by consent
Duane Morris's February 2026 client alert compresses the position into four instructions: keep AI transcription off by default, document consent before toggling it on, review the vendor's retention and training terms, and add attorney impressions to the record so work-product protection has something to attach to. It also lists twelve all-party-consent states, among them California, Florida, Illinois, Maryland, Pennsylvania and Washington, where recording without every participant's agreement is unlawful regardless of what the bar thinks.
The practical consequence is that the meeting bot cannot be a standing feature of your calendar. It is a per-meeting decision with a per-meeting consent record.
What the bar opinions actually say
| Source | Date | What it requires | Where it bites |
|---|---|---|---|
| NYC Bar Formal Opinion 2025-6 | 22 Dec 2025 | Client meetings: informed consent of client and all participants; no surreptitious recording; lawyer responsible for accuracy of summaries | Any client call with a notetaker joined |
| NYC Bar Formal Opinion 2026-2 | 5 Aug 2026 | Non-client conversations (opposing counsel, witnesses, experts): consent of all parties; secret recording is dishonest conduct under Rule 8.4; default is not to record | Negotiation calls and witness interviews |
| ABA Formal Opinion 512 | 29 Jul 2024 | Competence, confidentiality (informed consent where a self-learning tool may expose client data), supervision | Choice of tool and tier |
| UK Judiciary AI guidance | 31 Oct 2025 | No private information into public AI tools; judges answerable for all material in their name | Any hearing-related recording |
| CCBE generative-AI guide | 2 Oct 2025 | Inform clients about AI use and let them object; hallucination and data-retention risks | Engagement letters across Europe |
| BRAK Hinweise | Dec 2024 | Tools that store or forward client data for training breach § 43a Abs. 2 BRAO | German firms choosing consumer tiers |
| ÖRAK practice rules | 19 Mar 2026 | Client data stays on EU servers; disclose AI use; lawyer keeps responsibility; no training on client data | Austrian firms using US-hosted notetakers |
The state bar opinions page tracks the seventeen US jurisdictions with formal AI opinions as of June 2026; the New York City opinions are the only two that deal squarely with meeting recorders.
Which tool, which data path
The tool question is really a data-path question. The transcript of a privileged conversation is the most sensitive document your firm will create that week, and it is created by a third party.
| Tool | Data path (as of September 2026) | Suitable for privileged meetings? |
|---|---|---|
| Microsoft 365 Copilot meeting recap (Teams) | Under the Microsoft DPA with Enterprise Data Protection and Purview; does not train on prompts | Yes, once your tenant admin has configured recording, retention and Purview labels |
| Consumer Copilot app or Copilot Pro | Consumer terms, no DPA; "not suitable for confidential or legal work" per the same analysis | No |
| Zoom, Otter, Fireflies and similar standalone notetakers | Vendor policies contemplate processing for service improvement; terms vary by plan and change; the bot joins as a visible participant | Only on a business plan with training excluded in writing and a DPA; never by default |
| Uploading an audio file to Claude, ChatGPT or Gemini | Business, Team, Enterprise and API tiers do not train; consumer tiers train unless opted out; consumer Gemini chats may be read by human reviewers | Yes on business tiers; no on consumer tiers |
| On-device transcription (local model, no upload) | Nothing leaves the laptop | Yes, if the firm controls the device and the model output |
Two details that catch people. First, the notetaker that joins a Zoom call is a participant, and both NYC opinions treat it as one: everyone on the call must know it is there and agree. Second, Google's own privacy page for consumer Gemini says not to enter confidential information that you would not want a reviewer to see and keeps reviewed chats for up to three years. The Copilot page and the privacy-settings playbook walk through the settings for each vendor.
Where privilege is at risk
Three scenarios account for most of the exposure.
The consumer-tool transcript. In United States v. Heppner (S.D.N.Y., bench ruling 10 February 2026, opinion 17 February 2026), Judge Rakoff held that a defendant's research memos produced with consumer Claude were neither privileged nor work product because "Claude is not an attorney" and the terms permitted logging and disclosure, while signalling that privilege might attach where counsel directs the AI's use. A week later, Warner v. Gilbarco (E.D. Mich., February 2026) protected AI-assisted work product on the basis that AI programs are "tools, not persons". The two rulings are reconcilable: what counsel directs, on a confidential data path, looks like work product; what a client or lawyer runs through a consumer app with permissive terms does not. The privilege page and the ChatGPT privilege question go deeper.
The discoverable verbatim record. In re OpenAI Copyright Litigation (S.D.N.Y.) ordered production of 20 million de-identified ChatGPT logs, affirmed by Judge Stein on 5 January 2026 on the reasoning that users voluntarily submitted their communications to OpenAI. Vendor-held logs are electronically stored information. A transcript in a notetaker's cloud is no different. Reports in August 2026, which could not be independently verified, described a wiretap and biometric class action against Otter.ai being allowed into discovery and a notetaker transcript becoming central to a discrimination claim; treat them as a warning about the direction of travel rather than as settled authority.
The raw transcript with no lawyer's mind in it. Work product protects the lawyer's mental impressions. A verbatim transcript contains none. The Duane Morris advice to add attorney impressions is the practical fix: the file note, authored and signed by the lawyer, is the document you want to defend; the transcript is a temporary input.
Consent in Europe: data protection and criminal law
In the EU, the recording is processing of personal data of every participant, including the opposing side's witness and the expert. You need a lawful basis (consent or, more defensibly for the firm's own client, the contract), a data-processing agreement under Article 28 GDPR with the vendor, and an answer to where the audio and transcript are stored. The GDPR page has the checklist.
Criminal law sits on top. Germany's § 201 StGB makes recording the non-public spoken word without consent an offence; Austria's § 120 StGB penalises recording or passing on conversations in circumstances the participants did not agree to. Neither provision has an "it was only an AI notetaker" exception. Announce the recording at the start of the call and ask for agreement on the record; the Austria and Germany guidance page covers the professional rules on top of that.
The setup that works
- Policy first. Write the meeting-recording section of your firm AI policy: default off, per-meeting consent, approved tools by tier, retention period, who may export.
- Tenant configuration. If you are on Microsoft 365, have the admin set Teams recording and transcription to off by default, require an explicit start, apply a Purview retention label to transcripts and block external notetaker bots from joining tenant meetings.
- Consent script. Two sentences at the top of every recorded call: that an AI tool will transcribe, and that anyone may object, in which case it stays off. Record the answer in the transcript itself.
- Engagement-letter clause. Tell clients in the letter that meetings may be transcribed by AI with their consent and that transcripts are deleted after the file note is made; the CCBE guide asks for exactly this, and the client disclosure page has wording.
- Post-meeting routine. Within 24 hours, generate the file note with the prompt below, read it against the transcript, add impressions, save the note to the matter, delete the transcript.
- Audit. Once a quarter, list every recording in the vendor console and reconcile it against the file notes. Orphaned transcripts are the ones that hurt.
The prompt: transcript into file note and action list
Run this on a business-tier model with the transcript attached. Module 3 of the AI Lab for Lawyers course covers audio and transcript handling, including this workflow.
You are drafting an internal file note for a lawyer from a meeting transcript.
The note will be reviewed and signed by the lawyer. Work only from the
transcript; do not add facts, law or advice that is not in it.
MATTER: [matter name and number]
PARTICIPANTS: [names and roles; mark who is the client]
DATE: [date]
Produce, in this order:
1. FILE NOTE (max 350 words, past tense, third person): what was discussed,
decisions taken, positions stated by each side, facts the client
confirmed or corrected. Attribute statements to speakers. Where the
transcript is ambiguous or inaudible, write [unclear] and quote the
surrounding words.
2. ACTION LIST as a table: action | owner | due date if stated | source
(timestamp or speaker). Include only actions actually agreed in the
meeting. Do not infer actions.
3. OPEN QUESTIONS: points raised but not resolved, as short bullets.
4. FOR THE LAWYER TO COMPLETE: a heading followed by an empty block labelled
"Impressions and advice" that the lawyer will write. Leave it empty.
Do not summarise legal advice given in the meeting as if it were settled;
report it as "the lawyer indicated that ..." so the lawyer can check it.
Flag any passage where a participant mentioned a deadline, a limitation
period or a court date, with the exact words used.
The empty "Impressions and advice" block is deliberate. The lawyer fills it in by hand, which is what turns a transcript summary into work product. The same discipline applies to deposition and transcript summaries, where the stakes are higher and the transcripts longer.
Retention: decide before you press record
Anthropic's Compliance API keeps Cowork, Claude Code and Chrome session transcripts for six years by default for Enterprise customers; consumer Claude keeps training-enabled chats for up to five years; Gemini keeps consumer activity for 18 months and reviewed chats for three years. None of those figures is a good retention period for a privileged conversation. Set your own, short, and configure the vendor to honour it. The file note is the record; the transcript is scaffolding.
Where this leaves you
Meeting AI is worth having. A file note that is ready an hour after the call, with an action list and the exact words a client used about a deadline, is better than the note most lawyers write from memory the next morning. The conditions are not onerous: ask everyone, use a business tier, write the note yourself where it counts, and delete the rest. Firms that treat the notetaker as a default participant rather than a per-meeting choice are the ones the next bar opinion will be written about.
Frequently asked questions
Can lawyers use AI to record and transcribe client meetings?
Yes, with conditions. The New York City Bar's Formal Opinion 2025-6 requires the informed consent of the client and all other participants, prohibits surreptitious recording and treats not recording as the default. The tool must protect confidentiality, which means a business-grade data path with no model training, and the lawyer remains responsible for the accuracy of any summary.
Does an AI notetaker waive attorney-client privilege?
Not automatically, but it can. Privilege survives disclosure to agents who help the lawyer provide advice, and a February 2026 Michigan ruling treated AI programs as tools, not persons. A consumer tool whose terms allow logging, training and disclosure is harder to defend; Judge Rakoff refused privilege for consumer-chatbot memos in United States v. Heppner. Use business tiers and document that counsel directed the use.
Is Microsoft Copilot safe for legal meeting summaries?
Microsoft 365 Copilot under an enterprise agreement runs under Microsoft's data-processing terms with Enterprise Data Protection and Purview controls and does not train on prompts. The consumer Copilot app and Copilot Pro run under consumer terms without a data-processing agreement and are not suitable for confidential work. Check which one your firm actually licensed before enabling meeting recap.
How long should a law firm keep AI meeting transcripts?
Long enough to produce and check the file note, then delete unless the client asks otherwise or a litigation hold applies. A verbatim transcript is discoverable, may contain personal data of third parties and adds nothing to the file that the lawyer-authored note does not. Write the retention period into the AI policy and configure the vendor's retention setting to match.
Put this into practice
AI Lab for Lawyers: a live, hands-on course for practising lawyers
4 live sessions (8 hours), recordings included, taught by Dr. Niklas Schmidt of Wolf Theiss. Rated 4.7/5 from 381 ratings; alumni from Norton Rose Fulbright, Paul Hastings, Pillsbury and Taylor Wessing.
Related reading
Regulation & Ethics
Attorney–Client Privilege and AI: Heppner, Warner, the NYT Logs and Public Chatbots
Consumer chatbot memos are not privileged. US v. Heppner, Warner v. Gilbarco and the 20-million-log OpenAI order, plus rules that keep privilege intact.
10 min read
Questions
Does Using ChatGPT Waive Attorney–Client Privilege?
Using ChatGPT rarely waives privilege in the technical sense; the courts' answer is worse. In Heppner, consumer-chatbot memos were never privileged at all.
6 min read
Tools
Microsoft 365 Copilot for Lawyers: Word, Outlook and the Legal Agent
Microsoft 365 Copilot for lawyers as of September 2026. Why 76 percent of firms use it, what $30 a seat buys, the Word Legal Agent, Agent 365 and its limits.
8 min read
Regulation & Ethics
US State Bar AI Ethics Opinions: A 2026 State-by-State Guide
As of June 2026, 17 US jurisdictions plus the ABA have formal AI ethics opinions. None bans AI. They agree on verification, split on client consent and billing.
11 min read
Regulation & Ethics
GDPR and AI Tools for Lawyers: Lawful Basis, DPIAs and Vendor Terms
How GDPR applies when law firms use Claude, ChatGPT or Copilot: controller and processor roles, lawful basis, DPIA triggers, transfers and vendor terms.
11 min read
Use Cases
AI Deposition and Transcript Summaries: A Litigator's Workflow
AI deposition summaries work when every quotation is checked against the transcript, the tool runs on a business tier and output is anchored to page and line.
9 min read