NYC-BAR 2025-12-22

What does a New York lawyer have to do when AI tools (whether the lawyer's or the client's) record, transcribe, and summarize attorney-client conversations on Zoom, Teams, or similar platforms?

Short answer: The opinion concludes the lawyer must obtain client consent before AI records the call, must independently review any AI-generated transcript or summary for accuracy, must safeguard confidentiality and consider privilege risks, must train and supervise subordinates on AI use, and (where the client uses the client's own AI) should set expectations in the retainer agreement and warn of confidentiality and privilege risks.
Disclaimer: Advisory only. Not binding precedent.
About this page: The plain-English summary, reader guidance, and Q&A below were written by Ezel based on the official ethics opinion. The original opinion (linked at the bottom of this page, or PDF in the sidebar) is the authoritative source for any reliance.
View original ethics opinion (PDF)

NYC Bar Formal Opinion 2025-6: AI Recording, Transcription, and Summarization of Client Conversations

Short answer: The opinion concludes that a New York lawyer who uses an AI tool to record, transcribe, or summarize attorney-client conversations must obtain client consent before recording, must independently review the AI's transcripts and summaries for accuracy, must safeguard confidentiality and consider privilege implications of retained records, and must train and supervise subordinate lawyers and non-lawyers on AI use. Where the client (not the lawyer) selects the AI tool, the opinion identifies setting expectations in the retainer agreement and warning of confidentiality and privilege risks as the lawyer's available options.

Disclaimer: This is an advisory ethics opinion. Advisory opinions are not binding; they interpret the New York Rules of Professional Conduct and are persuasive authority. This summary is for research purposes only and is not legal advice. Verify current rules before acting on any specific guidance.

About this page: The plain-English summary and Q&A below were written by Ezel based on the official opinion. We do not reproduce the opinion text on this page; follow the linked source for the official text, which controls.

View original opinion

Plain-English summary

The opinion revisits and extends the committee's earlier Opinion 2003-02 (undisclosed taping) in light of AI tools that are increasingly default features of videoconferencing platforms (Zoom, Teams) and that go beyond audio recording to produce attributed transcripts and generative-AI summaries. The opinion divides the analysis between attorney-deployed AI and client-deployed AI.

On consent (attorney-deployed AI), the opinion concludes that even where the resulting summary might otherwise resemble a lawyer's note of the call, the existence of a voice recording and AI-generated transcript as predicate steps makes the AI process materially different from manual note-taking. The opinion holds that clients must be notified, and consent obtained, whenever a call is recorded by an AI-empowered system. The opinion grounds the consent requirement in Rule 8.4(c) (the duty against deception) and the duty of loyalty (Preamble [2]), tracking Opinion 2003-02's reasoning that participants speak differently when they know a verbatim record is being made.

On confidentiality and privilege (Rule 1.6), the opinion identifies retention as the central operational question. Per the opinion, attorneys must consider data storage location and duration, retrieval through discovery, whether the AI provider uses data for training, and whether the user can delete data. The opinion observes that preserved recordings can support an advice-of-counsel defense but can also expose unconsidered remarks; firms should set retention policies, disclose them in engagement letters, and follow them.

On competence (Rule 1.1), the opinion concludes the lawyer must independently verify the accuracy of any AI-generated transcript or summary before relying on it; AI work products must not be relied upon without review. The opinion also identifies a Rule 1.1 duty to acquire facility with the AI tool: understanding its operation, terms of service, default settings (especially default recording features), and security features.

On supervision (Rules 5.1, 5.3), the opinion concludes lawyers must train and supervise subordinates on AI use, building on the committee's Opinion 2024-5.

On client-deployed AI, the opinion observes the attorney's control is sharply reduced. Per the opinion, the lawyer cannot ensure accuracy, control retention, or guarantee the security features of a tool the client selected. The opinion identifies three options for the lawyer (none a complete solution): (1) ask that the conversation not be recorded; (2) include a retainer provision that any client-AI-generated record will not be deemed dispositive against the lawyer unless promptly shared so the lawyer can review; and (3) warn the client of the confidentiality and privilege risks.

The opinion also flags biometric-data laws (Illinois, Texas, Colorado, Maryland, Oregon, Washington) that apply to AI tools' use of "voiceprints"; New York currently does not have such a law, but the opinion notes pending legislation.

In practice

Under this opinion, a lawyer in New York who uses AI to record, transcribe, or summarize attorney-client conversations should: obtain client consent before recording (per call, or via an engagement-letter provision if recording is to be routine); verify all AI-generated transcripts and summaries promptly while details are fresh; set and follow a retention policy disclosed to clients; vet the AI provider's terms of service, training-data use, retention, and security; and train and supervise subordinate lawyers and non-lawyers on AI use.

Where the client uses the client's own AI, the opinion identifies the retainer agreement as the primary tool: agreement that the client will not use unapproved AI tools, or (failing that) that AI-generated records produced by the client's tool are not dispositive unless promptly shared for the lawyer's review.

Common questions

Q: Does Rule 8.4(c) bar a lawyer from using AI to record a client call without telling the client?

A: The opinion concludes that recording an attorney-client conversation with an AI tool requires client notification and consent. The opinion grounds the requirement in Rule 8.4(c) (deception) and the duty of loyalty under Preamble [2], following the reasoning of the committee's Opinion 2003-02. Per the opinion, the existence of a voice recording (even if transcribed only by AI and then deleted) makes the AI process materially different from manual note-taking.

Q: Can a lawyer rely on an AI-generated summary of a client call without reviewing the transcript?

A: The opinion concludes no. Per Rule 1.1, the lawyer must independently verify the accuracy of any AI-generated transcript or summary. The opinion warns against relying on AI summaries that may contain hallucinations or misattribute statements, especially given that summaries may be referenced years later.

Q: What does the lawyer have to know about the AI tool itself?

A: The opinion identifies several Rule 1.1 competence requirements: a non-technical understanding of how the tool works, awareness of the provider's terms of service, knowledge of default settings (including default recording features in videoconferencing platforms), and understanding of the tool's security features and limitations.

Q: If the client wants to use the client's own AI to record the call, what should the lawyer do?

A: The opinion identifies three options. The lawyer may ask that the conversation not be recorded; may include in the retainer agreement a provision that client-AI-generated records are not dispositive against the lawyer unless promptly shared for review; and should warn the client of the loss of confidentiality and privilege risks tied to a tool the lawyer cannot vet or control.

Q: Do other states' biometric-data laws affect attorneys using AI transcription?

A: The opinion observes that Illinois, Texas, Colorado, Maryland, Oregon, and Washington have biometric-data laws covering voiceprints, and that AI transcription tools use voiceprints to attribute statements. Per the opinion, AI recording without client knowledge and consent may violate such laws.

Q: What about default AI recording features in Zoom and Teams?

A: The opinion identifies the default-on recording features as a particular Rule 1.1 concern. Per the opinion, the lawyer should understand how to disable such features and what to do if a conversation is inadvertently recorded (the opinion notes deletion with client permission as one option).

Background and rules framework

The opinion interprets New York Rules 1.1 (competence, including technology competence under Comment 8), 1.6 (confidentiality), 5.1 and 5.3 (supervision of subordinate lawyers and non-lawyers), and 8.4(c) (prohibition on deception). The opinion also invokes Preamble [2]'s articulation of the loyalty duty. The opinion builds on the committee's Opinion 2003-02 (undisclosed taping) and Opinion 2024-5 (generative AI in practice), and references state biometric-data laws and one-party-consent vs. all-party-consent recording statutes.

Citations and references

Rules of Professional Conduct:

  • N.Y. RPC 1.1 (competence; Comment 8 on technology competence)
  • N.Y. RPC 1.6 (confidentiality)
  • N.Y. RPC 5.1, 5.3 (supervision of lawyers and non-lawyers)
  • N.Y. RPC 8.4(c) (deception, dishonesty, misrepresentation)
  • N.Y. Rules of Professional Conduct Preamble [2] (loyalty)

Statutes:

  • N.Y. Penal Law § 250.00 (eavesdropping; one-party-consent framework)
  • N.Y. C.P.L.R. § 4506 (admissibility of eavesdropping evidence)
  • State biometric-data laws (Illinois, Texas, Colorado, Maryland, Oregon, Washington), noted by jurisdiction
  • New York City Local Law 3 of 2021 (biometric data; noted as not applying to lawyers/law firms by its terms)

Cases:

  • People v. Badalamenti, 124 A.D.3d 672 (2d Dep't 2015), cited on the one-party-consent framework.
  • People v. Clark, 19 Misc. 3d 6 (App. Term, 2d Dep't 2008), cited on the same.
  • Rotonde v. Stewart Title Ins. Co., 86 Misc. 3d 1213(A) (Sup. Ct. Westchester Co. 2025), cited as one of several cases on AI hallucinations and sanctions.
  • Matter of Samuel, 82 Misc. 3d 616 (Surrogate's Ct., Kings Co. 2024), cited on the same.
  • Benjamin v. Costco Wholesale Corp., 779 F. Supp. 3d 341 (E.D.N.Y. 2025), cited on the same.
  • United States v. Cohen, 724 F. Supp. 3d 251 (S.D.N.Y. 2024), cited on the same.
  • Mata v. Avianca, Inc., 678 F. Supp. 3d 443 (S.D.N.Y. 2023), cited on the same.

Other opinions cited:

  • NYC Bar Op. 2024-5: ethical obligations of lawyers using generative AI in the practice of law.
  • NYC Bar Op. 2003-02: undisclosed taping of conversations by lawyers.
  • N.Y. State Bar Op. 1020 (2014): cloud storage and informed consent on confidentiality risks.
  • N.Y. State Bar Op. 328 (1974): historical position on undisclosed electronic recording.
  • N.Y. County Lawyers Op. 696 (1993): qualified historical permission to record with one-party consent.

See also

Source