Guides · For lawyers
AI note-takers in client meetings: what the ethics opinions and courts say.
General information, not legal advice. Ethics rules differ by jurisdiction, and most of the opinions below are advisory. Check the rules where you practice and where the conversation takes place.
An AI note-taker can record, transcribe and summarize a client meeting in minutes. It can also put the conversation on a vendor's servers, where it may be kept, used to train models or sought in litigation. Here is what bar opinions and courts have said so far, and what to check before you press record.
Two opinions deal with AI recording directly
Most ethics guidance on AI is general. As of October 2026, the only formal opinions we found that deal specifically with recording, transcribing and summarizing conversations come from the New York City Bar:
- Formal Opinion 2025-6 (December 22, 2025), conversations with clients. Tell clients and get their consent before an AI tool records them; the engagement letter can address it. Before using a tool, know where it stores the data and for how long, whether the data trains its models, whether it could be discoverable, and whether it can be deleted. Know how to switch off default recording. Review recordings, transcripts and summaries promptly yourself, including any legal analysis the tool produces, and train the staff who use it (NYC Bar).
- Formal Opinion 2026-2 (August 5, 2026), conversations with anyone else. Get the consent of every other party before recording, and make not recording the default unless there is a good reason. It gives separate guidance for co-counsel, prospective clients, opposing counsel, witnesses and investigators (NYC Bar).
Both apply New York's rules and are advisory. They are still the most detailed guidance available, and others are building on them: an article from the Illinois Supreme Court Commission on Professionalism applies 2025-6 to Illinois lawyers' duties (2Civility, May 2026).
The general rules on AI and recording still apply
ABA Formal Opinion 512 on generative AI
The ABA's 2024 opinion does not mention note-takers, but its rules reach any tool that receives client information (ABA Formal Opinion 512, July 2024):
- Before client information goes into a tool, assess the risk that it is disclosed or accessed, inside or outside the firm.
- If the tool could expose one client's information to others, for example by learning from it, get the client's informed consent first. Boilerplate in an engagement letter is not enough.
- Read the tool's terms of use and privacy policy, or have someone qualified read them.
- Vet the vendor as you would any outsourcing: its security, its confidentiality obligations, whether it keeps or claims rights in your data, and whether it will tell you about a breach or a demand for your data.
The older rules on recording
ABA Formal Opinion 01-422 (2001) says that recording without everyone's knowledge does not by itself breach the Model Rules where it is lawful, but it can where the law requires everyone's consent, and a lawyer may never falsely deny that a conversation is being recorded. It also says it is almost always advisable to tell clients. ABA Formal Opinion 498 (2021) calls recording client conversations on a meeting platform inadvisable without the client's consent, and advises switching off smart speakers and voice assistants during client discussions (ABA Formal Opinion 498). States differ: Texas has allowed lawyers to record without disclosure under conditions (Texas Opinion 575). Whether a recording is legal is a separate question, covered in recording consent laws by state.
State opinions on AI
California (revised May 2026), Florida, Texas, the District of Columbia, New Jersey, Pennsylvania and North Carolina have issued general guidance on AI. None addresses note-takers specifically, but they agree on the basics: understand how a tool collects, stores, uses and discloses data, and rely on its terms rather than its marketing. Several, including California and Florida, call for the client's informed consent before confidential information goes into a tool that poses a material risk (California, Florida 24-1, Texas 705, DC 388).
Privilege: what the courts have said so far
We found no decision holding that a lawyer's use of an AI note-taker waives privilege. The rulings so far involve other uses of AI, and they turn on what the provider may do with the conversation:
- United States v. Heppner (S.D.N.Y., February 2026). A criminal defendant's own exchanges with a consumer AI chatbot, made without his lawyers' direction, were not privileged. The AI is not a lawyer, which the court said settled the question on its own. It added that the exchanges were not confidential, because the provider's privacy policy let it collect inputs, train on them and disclose them to third parties, and that they were not made to get advice from counsel. It suggested the answer might differ if counsel had directed the use (memorandum).
- Warner v. Gilbarco (E.D. Mich., February 2026) and Morgan v. V2X (D. Colo., March 2026). Self-represented litigants' use of AI tools did not waive work product protection. In Morgan, the court's protective order allowed confidential material into an AI tool only if the provider was contractually barred from storing it or training on it, with a right to have it deleted (Warner, Morgan).
- In re Otter.AI Privacy Litigation (N.D. Cal., August 2026). At the pleading stage, the court let federal and California wiretap claims proceed against a note-taker company because the plaintiffs plausibly alleged it keeps recorded conversations and uses them to train its models. It said a software provider can be a third-party eavesdropper in those circumstances, where a provider that is merely the user's tool would not be. Claims that its speaker profiles are voiceprints under Illinois law also survived. These are allegations, not findings, and the company has answered the complaint (order).
The thread running through these: the more a provider may keep, use or disclose, the weaker the claim that the conversation stayed confidential. A tool that never gives anyone else a copy avoids the question. That is our reading, not a holding.
Voiceprints are biometric data in some states
A speaker-recognition feature that stores a voice profile to identify people later can create a voiceprint, and several states treat voiceprints as biometric identifiers:
- Illinois. The Biometric Information Privacy Act requires written notice and a written release before collecting a voiceprint, and a public retention and destruction policy. People can sue, for $1,000 per negligent violation and $5,000 per intentional or reckless one. The Act covers individuals, partnerships and corporations, so it reaches law firms (740 ILCS 14).
- Texas. Capturing a voiceprint for a commercial purpose requires notice and consent, and it must be destroyed within a year after the purpose ends. The Attorney General enforces it, with penalties up to $25,000 per violation (Bus. & Com. Code § 503.001).
- Washington. Enrolling a voiceprint in a database for a commercial purpose, which the law defines narrowly, requires notice and consent (RCW 19.375).
- Colorado. Voiceprints have been biometric identifiers under the state's privacy law since July 1, 2025 (HB24‑1130).
The NYC Bar's opinion also lists Maryland and Oregon. Note-taker companies are being sued over this: besides the Otter case, a complaint against Fireflies.AI alleges its speaker recognition created voiceprints of every participant in a meeting, including people without an account (Cruz v. Fireflies.AI, C.D. Ill., filed December 2025). In August 2026 the Seventh Circuit held that a device maker had not possessed or collected face templates under the Illinois Act when the templates stayed on users' own devices, beyond its control (G.T. v. Samsung). Whether that reasoning reaches voice, and whether a firm that uses such a tool is itself collecting voiceprints, is still open.
Labeling who spoke within a single recording, without keeping a profile, is less clearly covered, and the law is unsettled. If your practice touches these states, treat any feature that remembers voices as biometric collection: tell people, get written consent where the law requires it, set a retention policy, and delete voice profiles when the matter closes.
When the client is the one recording
Clients bring their own note-takers too. The NYC Bar suggests asking clients not to record without telling you, agreeing that output from a client's tool does not bind you unless they share it for your review, and warning them that recording can cost them confidentiality and privilege. ABA Formal Opinion 523 (May 2026) confirms that an engagement agreement may include the client's promise not to record lawyer-client communications, and that persistent breach can support withdrawal (ABA). Heppner shows the risk is real: there, the client's own use of a consumer AI tool is what lost the privilege.
Questions to ask before you use any recording tool
Drawn from the opinions above and the Illinois ARDC's guide to implementing AI (ARDC):
- Where is the audio processed and stored? On the device, or on the vendor's servers, and in which country?
- Does the vendor train on it? Recordings, transcripts or summaries, by default or unless you opt out?
- How long is it kept? Including after you delete it, and after the service ends.
- Who else can reach it? Every subprocessor, and the vendor's own staff.
- What happens in a breach or a subpoena? Will the vendor tell you, and does the contract say so?
- What do the terms say? The contract and privacy policy, not the marketing page.
- Which note-takers are already on? Know the ones built into your meeting platforms, and how to switch them off.
- Does it remember voices? If so, where are the voiceprints kept, and how are they deleted?
- Who checks the transcript? And where does it go in the client file, under which retention policy?
Engagement letter language for recording on a phone
The opinions point to the engagement letter as the place to set expectations. None of the guidance we read includes wording for transcription that runs on the device, so here is a starting point to adapt. It is a suggestion, not approved language.
- "With your permission, we may record our meetings on a firm phone so that our notes are complete and accurate. We will tell you before any recording starts, and you may ask us not to record at any time."
- "Recordings are transcribed by software on that phone. The audio is not sent to a transcription service, and it is not used to train any AI system."
- "A lawyer reviews each transcript. Transcripts are kept in your client file under our retention policy, and audio recordings are deleted [after review / when the matter closes]."
- "Please tell us before you record our conversations or use an AI tool on them, because doing so can affect confidentiality and privilege."
Make sure the words match the device. Use a phone the firm controls, protected by a passcode, and check its backup settings, because a cloud backup of the phone includes what its apps store.
Questions
Does using an AI note-taker waive attorney-client privilege?
As of October 2026 we found no decision holding that a lawyer's use of an AI note-taker waives privilege. Courts look at whether the conversation stayed confidential under the provider's terms and whether the provider acted as the lawyer's agent. In United States v. Heppner, a client's own use of a consumer AI tool was not privileged, partly because the provider's policy let it keep, train on and disclose what was typed.
Do I need my client's consent to record with an AI tool?
The NYC Bar's Formal Opinion 2025-6 says clients must be told and must consent before an AI tool records them. ABA Formal Opinion 512 requires informed consent before client information goes into a tool that could expose it. Separately, some states require every party's consent to any recording.
Is transcribing on the device safer than a cloud service?
It takes the vendor out of the conversation: no third party receives, keeps or trains on the audio. Your other duties stay the same, including consent, securing the device, retention and reviewing the transcript.
Do speaker labels create a voiceprint?
A feature that stores a voice profile to recognize a person later can create a voiceprint, which Illinois, Texas, Washington and Colorado treat as biometric data. Labeling speakers within one recording without keeping a profile is less clearly covered. The law is unsettled.
Where Interview fits
Interview records the conversation in the room and transcribes it on the phone itself. It does not join calls as a bot, and it sends no audio or transcript to us or to any transcription service, so for that step there are no vendor terms, retention or training to assess. The app sends anonymous usage counts, never audio or text, as its privacy policy describes. It remembers a voice only when you name a speaker or set up your own, keeps each voiceprint on the phone, and lets you delete it in Settings. Transcripts export to Word, PDF, Markdown or plain text for the file.
Related
Interview for lawyers · Recording consent laws by state · Interview vs Otter and Fireflies