AI Meeting Notes for Real Estate Closings: Confidentiality Beyond Attorney-Client Privilege

A residential purchase closing usually has a title agent, a lender rep, a buyer, a seller, two brokers, and — depending on the deal — a 1031 qualified intermediary, a condo board attorney, or a lender's outside counsel. A commercial closing multiplies that cast. Attorney-client privilege, in the strict sense, covers only a slice of what gets said on those calls. But the confidentiality obligations under Model Rule 1.6, plus contractual NDAs, RESPA-adjacent duties, GLBA safeguards on nonpublic personal information, and state consumer financial privacy laws, cover a great deal more.

That mismatch matters when a lawyer turns on an AI meeting notetaker. Privilege analysis asks whether a communication was confidential and made for legal advice. Confidentiality analysis asks a broader question: did the lawyer take reasonable steps to prevent unauthorized disclosure of any information relating to the representation? Real estate practice is where that broader duty bites hardest, because so much of what gets discussed on a closing call is not privileged at all — and yet its disclosure would still be a professional problem.

This article walks through where the exposure sits, what the recent guidance actually says, and how the architecture of an AI notetaker changes the risk calculation. If you want the product-level view, see Basil for Law.

Why Privilege Is the Wrong Frame for Closings

Privilege protects confidential communications between a lawyer and client for the purpose of legal advice. On a closing call, the lawyer speaks with the client — but also with the title company, the lender, opposing counsel, the escrow agent, and sometimes the broker. The moment a non-client, non-agent third party is on the line, privilege for that portion of the conversation usually evaporates under the classic third-party waiver rule, subject to narrow exceptions like the common-interest doctrine.

So the interesting question is not "is my closing call privileged?" It usually isn't, in whole. The interesting question is: what confidentiality duties do I still owe over the recording, the transcript, and the summary? Under ABA Model Rule 1.6, the duty of confidentiality extends to all information relating to the representation, whatever its source, and whether or not it would be privileged in litigation.

That covers the buyer's financing structure, the seller's tax posture, the fact that a 1031 exchange is running, the identity of a beneficial owner behind an LLC purchaser, mortgage payoff figures, price reductions negotiated off-market, and the fact of the transaction itself in some cases. Very little of that is privileged. Almost all of it is confidential.

The Categories of Sensitive Information on a Closing Call

It helps to sort what actually shows up on the recording, because the compliance analysis is different for each.

The point of the table is not that every category needs the same treatment. The point is that if an AI notetaker captures the call, the resulting artifacts touch every row. A single transcript will contain material governed by several overlapping regimes.

What the Recent Guidance Actually Says

ABA Formal Opinion 512, issued in July 2024, is the anchor. It tells lawyers using generative AI tools that they must understand how the tool handles inputs, whether prompts and outputs are retained or used to train models, and whether third parties can access the data. It ties this back to competence (Rule 1.1), confidentiality (Rule 1.6), supervision (Rules 5.1 and 5.3), and communication with the client (Rule 1.4).

Opinion 512 does not ban cloud AI. It does say that if a tool ingests client information into a system where the vendor can read it, retain it, or use it to train models, informed client consent is generally required — and in some configurations, the tool may not be usable at all. For a closing call, informed consent from the lawyer's own client does not solve the problem, because the recording also contains information belonging to non-clients (the counterparty, the lender, the title company) who never consented to anything.

New York's NYC Bar Formal Opinion 2025-6, issued in December 2025, extends this analysis specifically to AI meeting assistants and notetakers. It emphasizes that recording a meeting introduces a durable artifact — the transcript — that lives longer and travels further than the human memory of the call, and that lawyers must think about who else can see that artifact under the vendor's architecture.

The Third-Party Doctrine Problem

Even setting Rule 1.6 aside, there is a litigation problem. In United States v. Heppner (S.D.N.Y. Feb. 2026), Judge Rakoff held that a litigant's chats with a public AI platform were not privileged, drawing on the third-party doctrine: information voluntarily disclosed to a third-party service provider generally loses its confidentiality shield.

That reasoning is not directly about meeting notetakers, but the analogy is uncomfortable. If a lawyer runs a closing call through a cloud notetaker that stores the recording and transcript on vendor servers, a court asked to compel production may see a third-party custodian holding transactional records, not a privileged internal file. Combine that with Brewer v. Otter.ai (2025), which surfaced the question of what notetaker vendors do with recordings they receive, and the picture gets clearer: the vendor's role in the data flow is now a discovery target, not a background detail.

The safest posture, and the one Heppner's logic points toward, is architectural: don't create the third-party custodian in the first place.

Wire Fraud, Nonpublic Personal Information, and Practical Exposure

Real estate closings are the single most attacked practice area for business email compromise and wire fraud. The FBI IC3 annual reports consistently place real estate among the top loss categories, driven by attackers intercepting wire instructions in transit. A cloud transcript that captures a title agent reading out wire details, a payoff amount, a bank name, and a closing date is a target-rich file. If the vendor is breached, or if a subpoena reaches the vendor, that file is exposed in a way the parties never contemplated.

GLBA's Safeguards Rule, which reaches lawyers indirectly through their handling of information from financial institutions, expects reasonable administrative and technical safeguards over nonpublic personal information. State laws — New York's SHIELD Act, California's CCPA/CPRA, Illinois's PIPA — layer on additional obligations. A cloud AI notetaker with subprocessors in multiple jurisdictions makes the vendor-diligence question much harder to answer honestly on the client's outside-counsel security questionnaire.

The Common-Interest and Joint-Representation Wrinkles

Some closings involve genuine common-interest arrangements: buyer's counsel and lender's counsel sharing information under a common-interest agreement, or joint representation of spouses buying a home. In those cases, information shared inside the common-interest circle can retain privileged status, but only if the participants treat it as confidential.

Introducing a cloud AI notetaker — especially one the other side's counsel did not select or vet — is precisely the kind of act that can break the shared confidentiality on which common interest depends. In West Technology Group v. Sundstrom (D. Conn. 2024), the court underscored how easily privilege claims collapse when third-party tools enter the picture without careful attention to who has access. For closings, the takeaway is that the notetaker's data flow is part of the confidentiality analysis, not separate from it.

Consent to Record: State-by-State Reality

Recording a closing call implicates state wiretap and eavesdropping law. Roughly a dozen states — including California, Florida, Illinois, Massachusetts, Pennsylvania, and Washington — require all-party consent for the recording of a private communication. The rest use one-party consent, but professional-responsibility opinions in many jurisdictions still expect lawyers to disclose recording.

For a closing with participants in multiple states — common in commercial deals — the safest practice is to assume all-party consent applies and to log it. That means a spoken disclosure at the top of the call, a written follow-up in the closing checklist or engagement letter, and a durable record of who said yes. This is one place where an AI notetaker can help rather than hurt: a consent-log feature that captures the disclosure in the transcript itself is an audit-friendly artifact.

See our earlier discussion in consent, recording, and AI notetakers for a fuller state-by-state treatment.

Retention: The Question No One Wants to Answer

Every closing generates a file that lives for years — the closing binder, the title file, the attorney's own matter file. But the audio and transcript from the closing call are new categories of record. How long should they live?

State bar rules on file retention vary; many require file retention for five to seven years after the matter closes. Nothing requires a lawyer to retain a raw audio recording of a closing call, and there are strong reasons not to: the recording contains nonpublic personal information, the transcript may capture wire details, and a longer retention window means a longer subpoena window. A defensible retention policy for closing-call recordings often looks like: keep the summary in the matter file; delete the raw audio and full transcript after a short, defined period once the summary has been reviewed.

That policy is only executable if the tool actually lets the lawyer delete. Cloud notetakers vary widely; some retain copies in backups even after a user-facing delete. On-device tools sidestep the question by never uploading in the first place.

Vendor Diligence Questions Before You Turn a Notetaker On

Before a closing, and ideally before adopting any AI notetaker, the diligence checklist looks like this:

• Where does the audio go the moment the microphone captures it? On-device, or uploaded?
• Where does transcription happen? On the device, or on vendor servers?
• Who are the subprocessors? Are they listed? Do any of them process outside the US?
• Is my data used to train models, either now or under a future policy change?
• What is the vendor's response to a subpoena directed at my client's data?
• Can I sign a DPA? An NDA? Are those standard or exception?
• What happens on delete — is it real deletion, or a soft flag on a retained record?
• Does the vendor have SOC 2 Type II, and does the scope actually cover the AI processing pipeline?

The ABA Law Technology Today archives and state bar technology sections have useful vendor-diligence templates. For AI-specific issues, Opinion 512's competence framing gives the structure: understand the tool well enough to explain it to the client.

Practical Workflow for a Confidentiality-First Closing

Putting the pieces together, a workflow that respects confidentiality across the whole cast of a closing looks something like this:

1. Before the call, decide whether to record at all. For sensitive transactions, contemporaneous notes may be enough.
2. If recording, disclose at the top of the call and log the consent inside the transcript.
3. Use a tool whose architecture matches your confidentiality analysis. If the tool uploads, your Rule 1.6 analysis has to reach every subprocessor.
4. Label the transcript and summary as Privileged & Confidential — Attorney Work Product where appropriate, understanding that the label does not create privilege where none exists but signals intent.
5. Segregate nonpublic personal information. Do not paste wire instructions or SSNs into shared summaries.
6. Apply a short retention window to raw audio. Keep the reviewed summary; delete the rest.
7. Document the workflow in the engagement letter so the client understands what is captured and how.

How Basil Approaches This

Basil runs on-device on macOS using the Apple Neural Engine. Audio capture, transcription, and summarization all happen on the lawyer's own machine. Nothing is uploaded, there is no vendor server, and there are no subprocessors — Basil never receives user data, because there is no Basil server to receive it. That is an architectural fact, not a policy promise, and it changes the confidentiality analysis: there is no third-party custodian, no vendor subpoena target, and no cross-border processing question.

Computer mode captures Zoom, Teams, and Google Meet directly on the machine, so no bot joins the closing call and no participant sees an unexplained attendee. Basil signs DPAs and NDAs on request. The Basil for Law edition — privilege attestation, consent log, matter organization, and Privileged & Confidential labeling — is launching in August 2026 at $19.99/month or $199.99/year for solo attorneys, with a 3-day trial on monthly and a 7-day trial on annual. The general Basil app, with a 60 minutes per month free tier, is available today. Details at Basil for Law.

This article is for information only and is not legal advice.

Frequently asked questions

Is a real estate closing call privileged?

Usually only in part. Privilege covers confidential attorney-client communications for legal advice. On a closing call, non-client participants — title agents, lenders, opposing counsel, brokers — are generally on the line, which strips privilege from those portions under the third-party waiver rule. The broader confidentiality duty under Model Rule 1.6 still applies to everything relating to the representation.

Does ABA Formal Opinion 512 prohibit cloud AI notetakers for closings?

No, it does not prohibit them. Opinion 512 requires lawyers to understand how the tool handles data, whether prompts and outputs are retained or used for training, and who can access the information. For closings, informed consent from the lawyer's own client does not cover non-client participants whose data is also on the call, which narrows the space in which a cloud notetaker is defensible.

Can I record a closing call in a one-party consent state without telling everyone?

Legally, in one-party consent states, one participant's consent is enough for wiretap purposes. Ethically, most bar authorities expect lawyers to disclose recording. If any participant is in an all-party consent state — California, Florida, Illinois, Massachusetts, Pennsylvania, Washington, and others — the stricter rule generally governs. Best practice on multi-state calls is to disclose at the top and log consent.

How long should I keep the recording and transcript of a closing?

State bar rules typically require file retention for five to seven years after matter closure, but they do not require retention of raw audio or transcripts. Because those artifacts contain nonpublic personal information and can be reached by subpoena, many firms keep the reviewed summary in the matter file and delete raw audio and full transcripts after a short defined period.

What is the risk of using a cloud notetaker if the vendor is subpoenaed?

If the vendor holds recordings and transcripts on its servers, it becomes a third-party custodian that can be served directly. Judge Rakoff's reasoning in US v. Heppner, applying the third-party doctrine to AI platform data, illustrates why courts may treat vendor-held records as outside the privilege umbrella. An on-device architecture avoids creating that custodian.

Does labeling a transcript Privileged & Confidential make it privileged?

No. Labels do not create privilege where the underlying communication does not qualify. They can signal intent to treat the material as confidential and support work-product arguments where applicable, but the substantive analysis — client, legal advice, confidentiality maintained — controls.

Keep client conversations on your device

Basil transcribes and summarizes entirely on-device — no cloud, no bot, no server to subpoena. See Basil for Law → · Legal-tool reviews →

This article is for information only and is not legal advice.