ABA Formal Opinion 512 and AI Meeting Recording: Client Consent, Vendor Due Diligence, and Choosing a Notetaker

Published October 31, 2026

Published October 31, 2026 · 11 min read

If you are a lawyer evaluating an AI meeting notetaker for client calls, ABA Formal Opinion 512 is the controlling national ethics guidance. Issued July 29, 2024 by the ABA Standing Committee on Ethics and Professional Responsibility, it tells you three things about recording a client meeting with an AI notetaker: evaluate the tool's data handling under Rule 1.6 before client information goes in, obtain the client's informed consent before using any self-learning tool with client data, and perform vendor due diligence under Rule 5.3 — the same way you would for any cloud or outsourcing provider. Opinion 512 does not ban AI notetakers. It makes you responsible for understanding exactly where your client's words go. This guide walks through what the opinion actually requires, how it maps onto popular tools like Otter, Fireflies, and Zoom AI Companion, and why a fully on-device notetaker materially shrinks the compliance surface.

What ABA Formal Opinion 512 actually says

LeanLaw's rule-by-rule breakdown captures the core: Opinion 512, issued in July 2024, is the first national ethics framework for lawyers using generative AI — and its core requirement is simpler than the anxiety around it suggests: use AI competently, protect client confidentiality, and supervise what the tool produces. The opinion applies six existing Model Rules — as summarized by GC AI: competence (understand the tool's capabilities and limits), confidentiality (protect client data and obtain informed consent), communication (tell clients about AI use when it is material), candor (verify output and avoid hallucinated citations), supervision (set policies and train staff), and reasonable fees (do not bill clients to learn AI in general).

For meeting recording specifically, the controlling rule is confidentiality under Model Rule 1.6, together with Rule 5.3's vendor-supervision duty. There is no separate ABA opinion on AI notetakers — Laxis notes that as of October 2026, no ABA formal opinion deals specifically with meeting note takers, so Opinion 512 is read together with state-bar follow-ups like New York City Bar Formal Opinion 2025-6 (December 2025), sharpening the confidentiality analysis for cloud-based AI.

The 'self-learning tool' trigger that most notetakers hit

Opinion 512's biggest demand kicks in when a tool is "self-learning." The National Conference of Bar Examiners' Bar Examiner quotes the opinion directly: because many of today's self-learning GAI tools are designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client, a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool. The risk, the opinion warns, exists even if the tool is used exclusively by lawyers at the same firm.

The Legal Prompts' rule-by-rule analysis frames the definition this way: Some GAI tools are described as 'self-learning,' meaning they will learn from themselves as they cull more data. Most consumer cloud notetakers meet that definition because their terms permit training on user content. Importantly, AI For Lawyers Academy's explainer stresses that boilerplate provisions in an engagement letter do not provide the informed consent needed before inputting confidential client information into tools that may learn from it. You cannot bury a line in your retainer and call it done.

The vendor due diligence Opinion 512 requires

Rule 5.3 — your duty to supervise non-lawyer assistants — now reaches your AI vendors. Frankfurt Kurnit's technology law summary explains that the Opinion draws on earlier opinions on cloud computing and outsourcing services to confirm that a lawyer's obligation to ensure vendors or third party contractors will do the work capably and protect confidentiality apply also to third parties' use of GAI tools. In particular, lawyers should investigate the reliability, security measures, and policies of the GAI tool and ensure that the GAI tool is configured to protect the confidentiality and security of information, that the confidentiality obligation is enforceable, and that the lawyer will be notified in the event of a breach.

The Maryland State Bar Association's summary turns this into a usable checklist: lawyers should conduct reference checks, verify vendor credentials, understand security policies, ensure confidentiality agreements are in place, be familiar with hiring practices, check for conflicts of interest, and have access to legal recourse in case of violations. For an AI notetaker, that means answering, in writing: where does the audio go, who can access it, how long is it retained, is it used to train models, who are the subprocessors, and what happens on subpoena or breach.

How Opinion 512 maps onto popular AI notetakers

The table below compares the main AI notetaker architectures against the Opinion 512 duties a lawyer must satisfy before approving a tool.

Architecture	Where audio goes	Training on content?	Rule 1.6 posture	Rule 5.3 vendor review burden
Cloud notetaker with meeting bot (Otter, Fireflies)	Vendor cloud + subprocessors	Yes by default on consumer tiers	Self-learning: informed client consent required	High — DPA, SOC 2, retention, training opt-outs, breach notice
Platform-native AI (Zoom AI Companion, Teams Copilot)	Platform vendor cloud	Varies; Zoom reversed course after backlash; Microsoft commits no foundation-model training	Enterprise terms often needed to clear Rule 1.6	Medium–High — read platform's enterprise addendum
Browser or desktop notetaker without bot (Granola, cloud mode)	Vendor cloud (audio or transcript)	Depends on tier	Still self-learning unless contractually disabled	Medium — review terms carefully
On-device notetaker (Basil AI)	Stays on the lawyer's device	No — no vendor access to audio	No third-party disclosure to assess	Low — nothing to retain, train on, or breach

The Otter.ai cautionary tale

The clearest real-world warning is the In re Otter.AI Privacy Litigation, a consolidated class action in the Northern District of California. Sheppard Mullin's analysis reports that Otter's AI-powered notetaker joins virtual meetings and records and transcribes participants' speech in real time. They also allege that Otter keeps the meeting data and uses it to train its AI models without getting consent from non-subscribers. This, they allege, violates Illinois' biometric law (BIPA) as well as eavesdropping laws. Namely, the federal Electronic Communications Privacy Act (ECPA) and California's Invasion of Privacy Act (CIPA). The court rejected Otter's defense that its bot was an authorized participant: Otter argued that its bot was an authorized meeting participant and therefore a party to the conversation. The court disagreed because Otter uses the notetaker for its own commercial purpose: training AI.

Kilpatrick Townsend's note frames the lesson for buyers: This case underscores the legal risks of using AI transcription tools and training AI on user data without clear, informed consent. Otter.ai's privacy policy claims it trains on 'de-identified' recordings, but courts and regulators have been skeptical of such assurances. For a lawyer, that is doubly dangerous — not only is there regulatory exposure, but the vendor's conduct is exactly what Opinion 512 asks you to foresee during due diligence.

What Otter's own policy says

Otter.ai's privacy policy, as summarized by Voibe's 2026 breakdown, lists 'training our proprietary AI technology on de-identified audio recordings and on transcriptions' as a purpose of processing. Voibe notes that Otter documents no training toggle, so check your account's data settings and, if there's no control, object in writing under the privacy policy and keep the reply. That is a long way from the kind of enforceable, in-writing "no training" commitment Opinion 512's vendor-review framework contemplates.

What "informed consent" actually looks like

Informed consent under Opinion 512 is more than a signature. The Legal AI Governance compliance guide quotes the opinion directly: because many of today's self-learning GAI tools are designed so that their output could lead directly or indirectly to the disclosure of information relating to the representation of a client, a client's informed consent is required prior to inputting information relating to the representation into such a GAI tool. Opinion 512, at 7. In practice this means naming the tool, describing what the vendor does with the audio and transcript, and giving the client a realistic opportunity to say no — before the meeting, not after.

The Laxis guide highlights how NYC Bar Formal Opinion 2025-6 interacts with Opinion 512 specifically for notetakers: Formal Opinion 2025-6 takes on note takers directly: get the client's consent before recording, consider whether recording is tactically wise given confidentiality and privilege, and check the transcript or summary for accuracy if it might be relied on. The opinion also notes that deleting the audio afterwards doesn't remove the need for consent, because people speak differently when they know a recording is being made.

Attorney-client privilege: the waiver question

The hardest question is whether using an AI notetaker on a client call can waive privilege. The honest answer is: it depends on the architecture. The Minutes app analysis explains the risk vector: 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. Because privilege depends on the confidentiality of the communication, piping audio through a vendor that retains and trains on it is the fact pattern that waiver doctrine looks for.

A post-Heppner brief on AI tools and privilege puts the real downside bluntly: Practically, this means the most damaging consequence of consumer AI use may be waiver over the underlying attorney-client communications themselves, not merely loss of protection for the AI outputs. For deeper treatment, see our analysis of whether AI notetakers waive attorney-client privilege.

What in-house counsel and small firms should ask vendors

If your IT or security team has banned Otter and Fireflies, Opinion 512 is on your side — and it gives you a concrete checklist to bring to any replacement vendor. CourtBriefly's implementation checklist is a good starting point: For each tool, document whether inputs are retained, who can access them and whether they train the model. Classify each tool: approved for client information, approved for non-confidential use only, or prohibited.

• Where does the audio actually go? If it leaves the device, name every destination: primary cloud, backups, subprocessors.
• Who at the vendor can access it? Role, geography, access logs, incident playbook.
• Is content used to train models? Get a written, contractual "no training" commitment — not a toggle that can be flipped silently in a policy update.
• Retention schedule, in writing. Explicit, bounded, and tied to a defined deletion event.
• Security certifications. SOC 2 Type II at minimum; HITRUST if PHI could appear.
• Breach notification. Clear timelines and contact paths.
• Can the vendor be subpoenaed? If audio exists on their servers, assume yes.

For a broader tool-by-tool look, see our AI meeting notetaker comparison guide.

How Basil AI solves this: on-device vs. cloud

Opinion 512's confidentiality and vendor-review duties exist because, in a cloud tool, the vendor is a disclosure surface. Remove the vendor from the audio path and the analysis changes. Basil AI runs 100% on the lawyer's iPhone, iPad, or Mac. Recording, speaker diarization, transcription, and summarization all happen on-device using Apple's Speech framework and the Neural Engine. The audio file and transcript stay in the device's local storage, protected by hardware encryption described at Apple's privacy page.

What that means for Opinion 512 compliance:

• Rule 1.6: There is no self-learning cloud tool processing client information, so the opinion's informed-consent trigger for self-learning tools does not fire. Lawyers still owe clients disclosure of material AI use under Rule 1.4 and must comply with state recording-consent statutes.
• Rule 5.3: There is no vendor retaining, training on, or accessing audio. The due-diligence file collapses to the device OS and the local app — not a web of subprocessors.
• Privilege: No third party is introduced into the communication. The audio exists only where the lawyer put it.

For deeper reading, see our pieces on bot-free local AI meeting notes and the broader buyer's guide for lawyers on privilege waiver.

What this means for different legal roles

For solo and small-firm lawyers running client intake

• Pick a tool whose data flow you can describe in one sentence to a client.
• Script a standing consent moment at the top of the call — don't rely on the retainer.
• Keep the recording on the device until you decide to export, then export deliberately.

For in-house counsel with an IT ban on cloud notetakers

• Bring the Opinion 512 vendor-review framework to the vendor meeting as your checklist.
• Prefer on-device or private-tenant architectures that let security sign off without carve-outs.
• Document the classification in your AI inventory: approved for client information, non-confidential only, or prohibited.

For managing partners and GCs setting firm policy

• Treat Opinion 512's six duties as the spine of your AI policy, not a separate program.
• Keep a vendor file for every approved AI tool, including the retention and training commitments.
• Train lawyers and staff that boilerplate consent is not sufficient under the opinion.

Opinion 512 is permissive, not prohibitive

It is worth closing on tone. Dizon Law's rule-by-rule read gets this right: Opinion 512 is permissive in spirit. It does not tell lawyers to avoid generative AI; it tells them to use it the way they are already required to use every other tool, with competence, confidentiality, candor, and honest billing. The opinion assumes lawyers will use AI meeting notetakers. It just insists you know what the tool does with your client's voice before you press record.

Choosing a tool that was architected to make that answer short — "the audio never leaves the device" — is the fastest path to a defensible Opinion 512 posture.