NYC Bar Formal Opinions 2025-6 and 2026-2: What Every New York Lawyer Must Know Before Using an AI Notetaker with Clients

Published August 18, 2026

August 18, 2026 · 11 min read

If you are a New York attorney and you have ever let Otter, Fireflies, Read, or Zoom AI Companion join a call with a client, the ethics landscape you were operating in on December 21, 2025 is not the one you are operating in today. On December 22, 2025, the New York City Bar Association's Committee on Professional Ethics issued Formal Opinion 2025-6: Ethical Issues Affecting Use of AI to Record, Transcribe, and Summarize Conversations with Clients. In 2026, the Committee followed it with Formal Opinion 2026-2, which extends the same reasoning to calls with anyone who is not a client. Together, the two opinions are the most detailed ethics guidance any U.S. bar association has published on AI notetakers — and they read like a spec sheet for what a defensible tool looks like.

This piece walks through what the opinions actually require, what they explicitly do not require, and how the architecture of the tool you choose changes which duties you can discharge with a checkbox and which ones you cannot delegate at all.

What Opinion 2025-6 Actually Says

The core holding is unambiguous. Per the NYC Bar's own summary, the opinion concludes that an attorney should obtain client consent before recording the call, should consider whether recording, transcribing and summarizing is tactically well-advised, and should check the work product for accuracy if there is a possibility the transcription or summary will be preserved. If the client is the one using the AI tool, the lawyer should advise the client of the disadvantages of doing so.

Underneath that summary sit four specific duties, each rooted in the New York Rules of Professional Conduct:

• Rule 8.4(c) — deception. Recording without client notification and consent is treated as deceptive conduct. The opinion follows the reasoning of the committee's earlier Opinion 2003-02 on secret recording, holding that 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.
• Rule 1.1 — competence. The lawyer must independently verify the AI's transcripts and summaries. A summary that misattributes a client statement or drops a critical qualifier is not a technology problem the lawyer can outsource.
• Rule 1.6 — confidentiality. Cloud AI transcription services are third parties receiving privileged material. The lawyer must vet the vendor's data retention, training-data use, security posture, breach disclosure, and deletion practices, and must use reasonable contractual and technical protections.
• Rule 5.1 / 5.3 — supervision. Subordinate lawyers and non-lawyers who deploy AI notetakers must be trained and supervised on these obligations.

New York is a one-party consent state under wiretapping law, but as the Legal AI Governance tracker notes, the ethics rules impose a higher standard: secret recording violates Rule 8.4's prohibition on deception and the duty of loyalty regardless of what the wiretap statute permits.

What Opinion 2026-2 Adds for Non-Client Calls

Six months later, the Committee released Formal Opinion 2026-2, addressing the ethics of AI recording on calls between attorneys and people who are not clients — opposing counsel, witnesses, experts, negotiation counterparties, deposition participants, and internal-investigation interviewees. The Committee, chaired by Michael E. Salzman, explains that the opinion builds on Opinion 2024-5 (Generative AI in the Practice of Law) and Opinion 2025-6, and that the general principles set out in those earlier opinions apply well beyond the attorney-client relationship.

Opinion 2026-2 states two general principles that matter most for tool selection: as part of their duties under Rule 8.4, attorneys must not surreptitiously record conversations, including videoconferences, with any third parties, and must disclose the intention to record the conversation. The opinion also discusses specific practice contexts — negotiations, witness interviews, and adversarial calls — where the visible-bot-equals-notice framing many cloud tools rely on is legally untested.

How These Opinions Fit the National Landscape

The New York opinions did not appear in a vacuum. On the litigation side, Mayer Brown's June 2026 client alert on AI notetakers analyzed Warner v. Gilbarco, Inc. (E.D. Mich. Feb. 10, 2026), holding that disclosure to a third-party AI vendor does not automatically waive work-product protection — but that the analysis turns on the vendor's data-retention and third-party-sharing practices, making vendor due diligence essential.

In parallel, Bloomberg Law characterized the current moment as one where AI notetaking has outpaced the law, leaving compliance for wiretapping and ethics rules unresolved and raising a fresh set of discovery questions. And K&L Gates notes that under FRCP 26(b)(1), parties may obtain discovery of non-privileged AI-generated material that is relevant and proportional — and Magistrate Judge Ona Wang's order in In re OpenAI, Inc., Copyright Infringement Litigation compelling production of millions of GenAI logs demonstrates that courts will not treat AI output as categorically off-limits.

Cahill Gordon's June 12, 2026 alert on AI Notetakers in Corporate Meetings puts the two threads together: questions of privilege, discoverability, and the integrity of corporate records are real concerns for compliance and legal professionals evaluating AI notetakers, and early decisions on privilege and work-product protection for AI-generated transcripts are trending in different directions.

The Vendor-Vetting Problem Rule 1.6 Creates

Rule 1.6 is where cloud AI notetakers get uncomfortable. Treated as third-party vendors receiving privileged information, tools like Otter and Fireflies must be evaluated on retention defaults, training-data use, contractual non-training commitments, breach history, and DPA language. The problem is that the actual answers are often unfavorable.

Otter.ai's privacy policy grants broad rights over user-uploaded content, and tl;dv's roundup of AI meeting recorder lawsuits notes that Otter is facing four consolidated federal suits in In re Otter.ai Privacy Litigation alleging recording without the required consent, with a motion-to-dismiss hearing that will be the first federal test of whether decades-old wiretap statutes reach modern AI notetakers. Fireflies has separately been sued twice in Illinois under BIPA, with the second case (Fricker v. Fireflies.AI Corp., N.D. Ill.) filed in March 2026 and damages sought at $1,000 per negligent violation and $5,000 per reckless or intentional violation, plus attorneys' fees.

Under Opinion 2025-6, a New York lawyer using either tool has to document why the tool's retention default, training-data pipeline, and litigation posture meet Rule 1.6's reasonable-efforts standard — and has to update that documentation as the litigation posture changes.

Cloud AI vs. On-Device Transcription: A Side-by-Side Under the NYC Opinions

The single most useful thing you can do when reading the two opinions is to lay the tool architecture next to the rule it implicates. Here is that comparison for the three duties most affected by tool choice:

The pattern the table shows is important: on-device does not do the lawyer's job for them on consent or competence review. What it does is collapse the Rule 1.6 vendor analysis and the discovery-exposure surface to something the lawyer can actually reason about without reading a 40-page DPA.

What Happens When You Do Not Follow Opinion 2025-6

The consequences track two axes: professional discipline and litigation exposure. On the discipline side, Opinion 2025-6 is binding ethics guidance on NYC Bar members and persuasive authority for all New York practitioners. Failure to obtain client consent for AI recording is now a Rule 8.4 problem, not a matter of professional taste.

On the litigation side, the exposure runs through the vendor. If your cloud transcription vendor is breached, subpoenaed, or ordered to produce logs, the client's privileged conversation is in the production. Under Warner v. Gilbarco's framework, whether that production waives work-product protection turns on the vendor's data-retention and third-party-sharing practices — facts largely outside the lawyer's control at the moment of the meeting.

The Retainer Letter Fix (and Its Limits)

Several commentators, including the Ezel Legal Ethics tracker, note that Opinion 2025-6 permits attorneys to obtain consent through an engagement-letter provision if AI recording is to be routine, rather than getting per-call consent. That is helpful for practices with high-volume client interaction, but it does not discharge the confidentiality or competence duties. A routine-consent clause does not tell you what happens when a specific conversation touches an especially sensitive matter — a settlement position, a threatened piece of litigation, an internal investigation — and Opinion 2025-6 makes clear that the lawyer must still consider whether recording is tactically well-advised in the circumstances.

The retainer clause also does not cover the non-client side of the ledger that Opinion 2026-2 addresses. A client's consent to your AI notetaker on your calls with them does not give you permission to run the same tool on your deposition prep call with a fact witness.

Practical Checklist for New York Firms

Based on the two opinions and the surrounding case law and commentary, a defensible AI-notetaker program in New York looks roughly like this:

1. Written consent policy. Bake it into the engagement letter and reinforce it verbally at the start of any recorded call. For non-client calls, obtain disclosure and consent at the top of the call.
2. Documented vendor evaluation. If you use a cloud tool, keep a memo on file that walks through retention, training-data commitments, DPA language, breach history, and litigation posture — and update it when facts change (as in In re Otter.ai Privacy Litigation).
3. Human accuracy review. Every AI-generated transcript and summary is reviewed against the audio before it is filed, sent, or used as the basis for advice.
4. Retention schedule. Written policy on how long AI transcripts, summaries, and audio are kept, and how they are destroyed. This is what makes work-product analysis tractable if the file is later subpoenaed.
5. Prefer on-device processing for the most sensitive matters. Where the conversation touches litigation strategy, MNPI, or an internal investigation, an on-device tool eliminates the vendor from the Rule 1.6 analysis entirely.

How Basil AI Solves This

Basil AI is a fully on-device AI meeting notetaker for iPhone, iPad, and Mac. It uses Apple's Speech Recognition framework and the Apple Neural Engine to transcribe audio locally — the same on-device processing philosophy Apple describes in its Privacy overview. No bot joins the meeting. No audio is uploaded to a Basil server. No transcript sits in a vendor cloud.

For a New York attorney trying to discharge Opinion 2025-6, that architecture does specific work:

• Rule 1.6 vendor-vetting collapses. There is no third party receiving the audio; the confidentiality analysis reduces to how the attorney's own device is secured.
• Training-data pipeline is a non-issue. The audio never reaches any pipeline — ours or anyone else's.
• Discovery exposure narrows to the device. The only copy of the transcript is the one the lawyer chose to save.

What Basil AI does not do — and what no notetaker can do — is obtain your client's consent, perform your accuracy review, or write your retention policy. Those remain the lawyer's professional obligations. "On-device" is an architecture fact, not an ethics safe harbor. For deeper reading on how the architecture affects related risks, see our analysis of AI notetakers, board meetings, discovery, and privilege, our guide to AI meeting notes for asset managers and MNPI, and our field guide to bot vs. botless AI notetakers for client-facing meetings.

The Direction of Ethics Guidance in 2026

Two other data points suggest the NYC Bar's approach is the direction of travel. Norton Rose Fulbright's Inside Tech Law analysis confirms that Opinion 2025-6 focuses on both the scenario where the lawyer is doing the AI recording and the scenario where the client is doing it — and imposes obligations in both directions. And the ABA's SciTech Lawyer coverage of AI in e-discovery emphasizes that attorneys must remain diligent in monitoring the accuracy and performance of AI tools to avoid hallucinations and biases — the same Rule 1.1 point Opinion 2025-6 makes for AI notetakers specifically.

The likely trajectory is that other state bars — California, Florida, Texas, Illinois — will issue analogous opinions over the next 12–18 months. New York's version will be the template; the vendor-vetting and retention questions are the same everywhere. The architectural question — where the audio actually goes — will be the shortest path to a defensible answer regardless of which state's rules apply.

The Bottom Line

The lawyers most exposed under Opinion 2025-6 are not the ones who read the opinion and pushed pause. They are the ones who inherited an Otter or Fireflies deployment from a firm-wide IT decision made in 2024, never renegotiated the DPA, and cannot answer — if asked in a disciplinary or discovery setting — where their client's audio is right now. On-device transcription does not answer every question the opinion raises. It answers the one question that is hardest to answer any other way.

Take Meeting Notes Without Handing Client Audio to a Third Party

Basil AI transcribes on your iPhone, iPad, or Mac. No bot in the meeting. No cloud copy of the conversation. No training-data pipeline.