AI Transcription for Immigration Lawyers: Protecting Client Confidentiality in a High-Stakes Practice

Immigration lawyers hear things almost no other practitioner hears. In a single intake you may learn that a client entered without inspection twelve years ago, that a spouse has a prior removal order under a different name, that an asylum applicant survived torture, or that a beneficiary's employer is quietly out of status on payroll taxes. These are not abstract confidences. They are facts that, if disclosed to the wrong system, could feed a government database, surface in a Freedom of Information Act response, or reappear during a stokes interview.

At the same time, immigration is a document- and narrative-heavy practice. Consultations run long. Country-conditions calls, preparation sessions for I-589 declarations, and Cancellation of Removal hardship interviews generate hours of material that has to be turned into usable notes, declarations, and briefs. AI transcription is an obvious productivity win — but only if the tool respects the confidentiality obligations that already govern the file.

This article walks through what those obligations actually require in 2026, what the recent authorities say about AI and confidentiality, and how immigration attorneys should evaluate an AI notetaker before turning it on during a client call.

Why Immigration Files Deserve Heightened Care

Every attorney-client relationship is covered by ABA Model Rule 1.6, which prohibits revealing information relating to the representation without informed consent and requires reasonable efforts to prevent inadvertent or unauthorized disclosure. Immigration practice layers on several extra pressures:

• Adverse-party interest is a government. USCIS, ICE, EOIR, and the Department of State can and do request records, cross-reference statements, and initiate NTAs. Any storage system that could be compelled or breached is a live risk to the client.
• Statements are portable across proceedings. An offhand admission in an intake — a date of entry, a prior name, an unauthorized work stint — can materially change removability or eligibility. Transcripts do not soften over time.
• Third-party family members are often present. A U-visa petitioner's mother, an unaccompanied minor's cousin, a derivative spouse — all may speak on the recording. Consent has to reach them, not just the client.
• Interpreters are common. Under Rule 1.6, an interpreter is treated as an agent of the lawyer, but if the interpreter's audio is streamed to a third-party AI vendor, the analysis is different.

The through-line is simple: the more sensitive the underlying facts, the more the architecture of your AI tools matters, not just the vendor's promises.

What ABA Formal Opinion 512 Actually Requires

In July 2024, the ABA Standing Committee on Ethics and Professional Responsibility issued Formal Opinion 512 on generative AI. The opinion is now the baseline reference for any lawyer using AI on client matters. For immigration lawyers, four points matter most:

1. Competence (Rule 1.1) requires understanding the tool. That includes where audio goes, how long it is retained, whether it is used to train models, and who at the vendor can access it.
2. Confidentiality (Rule 1.6) requires informed client consent before inputting information relating to the representation into a self-learning or third-party-hosted AI tool. A generic engagement-letter clause is not always enough.
3. Supervision (Rules 5.1 and 5.3) extends to non-lawyer AI tools the same way it extends to paralegals and vendors.
4. Fees (Rule 1.5) cannot be inflated by billing hours the AI saved.

Opinion 512 does not ban cloud AI. It does say the lawyer has to actually understand what the vendor does with the data. That is a meaningful hurdle for immigration counsel whose files touch government-adverse facts.

NYC Bar Opinion 2025-6 and the Consent Question

In December 2025, the New York City Bar issued Formal Opinion 2025-6, which sharpened the informed-consent analysis for AI notetakers specifically. The opinion treats an AI transcription service that stores or processes client audio on third-party infrastructure as a disclosure that generally requires the client's informed consent, and it emphasizes that consent must be specific enough for the client to understand where the audio is going.

For an immigration client — often a non-native English speaker, sometimes with limited experience with U.S. legal norms — informed consent is not a checkbox. It is a conversation. And it is materially easier to have that conversation when you can honestly say the recording never leaves the laptop.

The Heppner Warning: AI Chats Are Not Automatically Privileged

In February 2026, Judge Rakoff issued a decision in United States v. Heppner (S.D.N.Y.) holding that a litigant's exchanges with a public AI chatbot were not protected by the attorney-client privilege. The court analogized the disclosure to the classic third-party doctrine: once a communication is shared with a third-party service provider that retains and processes it, the confidentiality assumption that undergirds privilege is weakened.

Heppner involved a party, not a lawyer, and it did not address transcription tools directly. But the reasoning is exactly what any immigration lawyer should be worried about: if the audio of an asylum intake is streamed to a cloud model that retains it, an aggressive opposing party could argue that the intake — or the derivative work product — was disclosed to a third party. The Heppner analysis is uncomfortable enough that the safest posture is to keep the recording off third-party servers in the first place.

Bot-Based Notetakers and the Consent Trap

A separate risk vector is the bot-based notetaker that dials into Zoom, Google Meet, or Microsoft Teams as a visible participant. In Brewer v. Otter.ai (N.D. Cal. 2025), plaintiffs alleged that a widely used AI meeting assistant recorded conversations without adequate consent from all participants, raising claims under state wiretap statutes. The case underlines a practical point that matters in immigration practice: if a bot named "Notetaker" pops into a call with a derivative spouse in a two-party-consent state, and no one affirmatively agrees to be recorded, the lawyer inherits the problem.

Compare that to West Technology Group v. Sundstrom (D. Conn. 2024), where the court examined the interplay between recorded business communications and confidentiality expectations. The lesson across both cases is the same: consent theater — a bot that announces itself but is easy to miss — is thinner protection than most lawyers assume.

What to Look For in an AI Transcription Tool

Immigration lawyers evaluating an AI notetaker should ask concrete architectural questions, not marketing questions. The following table maps the questions to what a defensible answer looks like.

For immigration counsel, the last three rows are load-bearing. A tool that cannot answer them cleanly is a tool that will eventually require an uncomfortable conversation with a client — or with a bar counsel.

Practical Workflow: Intake to Declaration

Assume you are preparing a client for an affirmative asylum interview. A defensible workflow looks like this:

1. Engagement letter discloses that the firm uses on-device AI transcription and explains, in plain language, that the audio does not leave the lawyer's computer. If the tool is cloud-based, disclose that instead and obtain specific written consent consistent with Opinion 512 and NYC Bar 2025-6.
2. Verbal consent at the start of each session, captured on the recording itself and logged. Where family members or interpreters are present, obtain their consent too. In two-party-consent jurisdictions, this is not optional.
3. Label the transcript as Privileged and Confidential — Attorney Work Product at the top of the file. If your tool applies that label automatically per matter, use it.
4. Segregate the country-conditions research from the client narrative. Public research can go through more permissive AI tools; the client's own words should not.
5. When drafting the I-589 declaration, work from the transcript locally. Do not paste the client's narrative into a public chatbot to "clean it up." That is precisely the scenario Heppner should make you nervous about.
6. Retention. Keep the audio only as long as the matter file requires under your jurisdiction's rules and your engagement letter, then delete deliberately.

Special Considerations for Removal Defense

Removal defense adds a layer. The client is often detained, the phone system is monitored, and the risk of a statement resurfacing at a master or individual hearing is real. Two rules of thumb:

• Never record a detention-facility call with a third-party cloud tool. Facility calls are typically monitored to begin with, and layering a cloud vendor onto a monitored line multiplies the disclosure surface without adding privilege.
• Use on-device transcription for prep sessions in your office or over video, where the confidentiality assumption is intact and you control the recording end-to-end.

For a broader treatment of privilege architecture, see our companion piece on Basil for Law and the case for local-first AI.

What Not to Do

• Do not paste a client's asylum narrative into a public chatbot for "help with grammar." That is the Heppner fact pattern.
• Do not rely on a bot-joins-the-call notetaker for sensitive intakes without explicit, per-participant consent — Brewer is a warning shot.
• Do not assume a vendor DPA cures a confidentiality problem. A DPA allocates liability; it does not un-disclose the client's story.
• Do not describe any AI tool to a client as making the conversation "privileged" or "protected." Privilege is a legal conclusion the court will reach or not reach. Architecture reduces risk; it does not guarantee outcomes.

How Basil Approaches This

Basil is an on-device AI meeting notetaker built for attorneys. Audio capture, transcription, and summarization all run locally on the Apple Neural Engine. Nothing is uploaded, there is no Basil server, and there are no subprocessors — because there is no cloud pipeline to have subprocessors in the first place. For virtual meetings on macOS, Basil captures system audio locally, so no bot joins the Zoom, Teams, or Meet call as a visible participant.

The Basil for Law edition launches in August 2026 and adds features aimed specifically at practice needs: a privilege attestation on every session, a consent log tied to the matter, matter-based organization, and automatic Privileged and Confidential labeling on exports. Pricing for the solo tier is $19.99/month or $199.99/year, with a 3-day trial on the monthly plan and a 7-day trial on the annual plan. The general Basil app, with a 60-minute monthly free tier, is available today. Basil signs DPAs and NDAs on request.

None of this makes a communication privileged as a matter of law. What it does is remove the third-party disclosure step that recent authorities keep flagging as the weak link — which, for an immigration practice, is exactly the step worth removing.

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

Frequently asked questions

Can I use a cloud-based AI transcription tool for immigration intakes if the vendor signs a DPA?

A DPA allocates contractual liability but does not undo the underlying disclosure. Under ABA Formal Opinion 512 and NYC Bar Opinion 2025-6, you still need to understand where the audio goes, how it is retained, and whether it is used for training, and you generally need informed client consent specific enough for the client to understand the data flow. For immigration files, where the adverse party is often a government, on-device processing removes that disclosure step entirely.

Do I need consent from an interpreter or family member on the call?

Yes. Interpreters are typically treated as agents of the lawyer for confidentiality purposes, but if a third-party AI vendor receives the audio, the analysis changes. Derivative spouses, parents of unaccompanied minors, and other family members on the call are separate participants whose consent you should obtain, particularly in two-party-consent jurisdictions.

Does the Heppner decision mean I cannot use AI at all on client matters?

No. Heppner held that a litigant's own exchanges with a public AI chatbot were not privileged, drawing on third-party doctrine reasoning. It does not prohibit lawyers from using AI. It does strongly suggest that keeping client audio and text off third-party servers is a safer posture, which is what on-device tools are designed to do.

What is the risk with bot-based notetakers that join Zoom or Teams?

Brewer v. Otter.ai alleged inadequate consent from meeting participants under state wiretap statutes. A visible bot named 'Notetaker' is easy to overlook, especially for a client on a first consultation. On-device system-audio capture avoids the bot-in-the-meeting scenario and lets the lawyer control the consent workflow directly.

Should I mention AI transcription in my engagement letter?

Yes. Disclose in plain language which AI tools the firm uses, whether audio is processed on-device or in the cloud, and how long recordings are retained. If the tool is cloud-based, that disclosure should be specific enough to support informed consent under Opinion 512 and NYC Bar 2025-6. An on-device tool makes that disclosure much shorter and easier for the client to understand.

When will Basil for Law be available?

The Basil for Law edition, which adds privilege attestation, a consent log, matter organization, and Privileged and Confidential labeling, is scheduled to launch in August 2026 at $19.99/month or $199.99/year for the solo tier. The general Basil app with a 60-minute monthly free tier is available today, and Basil signs DPAs and NDAs on request.

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.