Are AI Meeting Transcripts Discoverable in Court? The Litigation-Hold Trap Hiding in Your Zoom Notes

Yes — AI-generated meeting transcripts are electronically stored information (ESI) and they are discoverable in U.S. civil litigation. The instant your company reasonably anticipates a dispute, every Otter summary, Fireflies recap, Zoom AI Companion note, and Microsoft Copilot meeting digest in your environment becomes evidence you are legally obligated to preserve. Mishandle that obligation and the sanctions menu under Federal Rule of Civil Procedure 37(e) ranges from "curative measures" all the way to adverse-inference instructions and default judgment.

The trap is that most employees do not think of the AI bot in the corner of the Zoom window as a document custodian. It is. And a wave of 2026 court guidance — together with the consolidated In re Otter.AI Privacy Litigation, 5:25-cv-06911 in the Northern District of California — has made it clear that opposing counsel knows exactly where to look.

Why AI transcripts qualify as ESI

The duty to preserve ESI is foundational to U.S. civil procedure. As New Jersey commercial litigator Deanna Koestel explained in a March 2026 analysis for NJBIZ, AI-generated meeting transcripts are electronically stored information (ESI), and once litigation is reasonably anticipated, they're subject to the same preservation rules as emails, contracts, and other business records. If a transcript is relevant, accessible, and within your company's control, it's likely discoverable — the same principles that apply to email apply to AI-generated transcripts.

This is not an academic point. AI-generated ESI — especially from notetakers, meeting summaries, auto-drafted emails, and chat assistants — is becoming a core discovery battlefield in employment cases, according to a March 2026 client alert from Fisher Phillips. The firm cites two federal employment cases decided in early 2026 where AI-generated materials were directly at issue.

What FRCP 37(e) actually says — and how it punishes deletion

Rule 37(e) governs what happens when ESI "that should have been preserved in the anticipation or conduct of litigation is lost." Per the rule's structure, if the loss prejudiced another party, the court "may order measures no greater than necessary to cure the prejudice." If the court goes further and finds the party "acted with the intent to deprive another party of the information," the sanctions get severe: presuming the lost information was unfavorable, instructing the jury that it may or must presume the information was unfavorable, dismissing the action, or entering default judgment.

The intent standard matters. The Second Circuit's decision in Hoffer v. Tellone held in February 2025 that to impose sanctions for spoliation under Federal Rule of Procedure 37(e)(2), the moving party must show, by a preponderance of the evidence, that the accused party acted with the "intent to deprive" another party of lost electronically stored information — mere negligence will not suffice. That sounds like a defendant-friendly standard, but as Logikcull's guide to revised Rule 37(e) notes, severe sanctions could only be granted by judges after a finding that the guilty party "acted with the intent to deprive", and selective deletion or post-litigation purges of transcripts are exactly the patterns that prove intent.

The Otter.ai class action: a preview of vendor-side exposure

The most consequential AI-notetaker case currently before a U.S. court is In re Otter.AI Privacy Litigation, consolidated before Judge Eumi K. Lee in the Northern District of California. In Brewer v. Otter.ai Inc., No. 5:25-cv-06911 (N.D. Cal. Aug. 15, 2025), the complaint alleges that Otter automatically joins Google Meet, Zoom, and Microsoft Teams meetings and records, accesses, reads, and learns the contents of conversations involving non-users without their consent, in violation of the federal Electronic Communications Privacy Act (ECPA). According to the allegations, Otter transmits call content to its servers in real time, uses participant conversations to train its machine-learning models, retains recordings indefinitely, and provides no meaningful disclosure to non-Otter participants.

The discovery implication is straightforward: if recordings live on a vendor's servers indefinitely, they are reachable by third-party subpoena even after the customer believes they have been deleted. As Robinson+Cole's Data Privacy + Cybersecurity Insider documented in December 2025, follow-on suits including Winston v. Otter.ai Inc. allege that Otter not only transcribes and stores meeting content but also sends follow-up emails, including partial transcripts and screenshots, to all invitees, even those who never joined the meeting — and that Otter's default settings provide no disclosure to non-user participants.

Those follow-up emails are themselves discoverable ESI sitting in dozens of unrelated inboxes — copies you never agreed to create and cannot recall.

Cloud notetakers vs. on-device transcription: a discovery-risk comparison

The architecture of your transcription tool determines who controls the ESI, how long it persists, and who can be subpoenaed to produce it. Here is a side-by-side view of the discovery surface for cloud notetakers versus on-device transcription.

Privilege waiver: when the bot becomes the "outsider"

For lawyers, the discoverability problem is compounded by a privilege problem. Mayer Brown's June 2026 analysis warns that for meetings involving privileged communications, organizations should evaluate whether the use of third-party AI notetakers creates an unacceptable risk of privilege waiver, particularly given that AI may indiscriminately capture privileged information alongside non-privileged business purpose communications — and this risk is no longer theoretical.

The case driving that warning is United States v. Heppner, 25 CR. 503 (JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026), where the court refused to extend attorney-client privilege to materials prepared with a consumer-grade generative AI tool. As New England Biz Law Update summarized, in Warner v. Gilbarco, Inc., a federal court in Michigan found that a plaintiff's use of ChatGPT to help prepare her pro se case was protected as work product and not discoverable, while in United States v. Heppner, a federal court in New York reached the opposite conclusion, holding that AI conversations generated by a criminal defendant were not protected and could be used as evidence. The split is unresolved — but the risk of an adverse ruling is now well-documented.

The New York City Bar Association's Professional Ethics Committee captured this in Formal Opinion 2025-6, which advises attorneys that AI-generated transcripts and summaries may be discoverable in litigation, and privileged communications could be inadvertently compromised. Our deep dive on avoiding privilege waiver with AI notetakers walks through the buyer-side checklist.

The "silent ESI" problem: chat, summaries, and biometric voiceprints

AI transcripts are not the only AI ESI in a modern meeting. Koestel's NJBIZ column flags a frequently overlooked source: chat messages exchanged during a Zoom or Teams meeting are a separate evidentiary record that can be just as damaging as anything said out loud — and Zoom automatically saves in-meeting chat logs, including messages sent privately between participants, to the host's local device or cloud account.

Then there are voiceprints. Per a May 2026 DataGrail analysis, AI notetakers that use speaker identification features can capture voiceprints, which trigger biometric privacy laws — and states like Colorado, Illinois, Texas, and Washington require consent prior to collecting biometric information and impose restrictions on its retention and disclosure. The Otter consolidated docket already includes a BIPA-specific theory in Walker v. Otter.ai Inc., No. 5:25-cv-07187 (N.D. Cal. Aug. 26, 2025), which alleges that Otter's software collects and uses "voiceprints" (unique biometric data) from meeting audio without notifying participants or obtaining the explicit, written consent required by the Illinois Biometric Information Privacy Act (BIPA).

Why employers carry the bag

If you are an in-house lawyer or HR leader, the lawsuit and the litigation hold tend to land in your inbox at the same time. HR Executive's April 2026 analysis of the Otter case notes that In re Otter.AI Privacy Litigation, now a consolidated case before Judge Eumi K. Lee in the U.S. District Court for the Northern District of California, alleges that Otter.ai's notetaking tools recorded private conversations without the consent of all participants and used those recordings to train its AI models without adequate disclosure — and employment attorneys say the case is already signaling where liability could land for employers.

The compounding international layer: recordings processed by U.S.-based vendors must comply with international transfer mechanisms such as Standard Contractual Clauses, and beginning in August 2026, the EU AI Act introduces a separate layer of obligation — AI systems used for worker monitoring and management may be classified as high-risk, a category that could encompass tools offering sentiment analytics or productivity scoring alongside transcription. We covered the EU AI Act timeline in our earlier piece on August 2026 compliance obligations.

How Basil AI solves this: on-device ESI you actually control

Basil AI runs transcription entirely on the device using Apple's Speech Recognition and the Apple Neural Engine. There is no upload, no cloud copy, and no vendor account that can be subpoenaed for your transcripts. From an e-discovery perspective, that changes three things:

1. The custodian is you, not a SaaS vendor

Your IT and legal teams already know how to issue litigation holds across employee laptops and iPhones — that is what corporate MDM is for. They do not have a workflow for subpoenaing a third party they did not contract with, which is what every "OtterPilot joined the call" scenario forces them into. With on-device transcripts, the hold lands on a custodian device the company controls.

2. Retention is a setting you own

Cloud notetakers default to indefinite retention. Basil AI transcripts live in your local storage and your Apple Notes via iCloud — you set the retention, you delete on schedule, and when a hold attaches, you suspend that schedule on the custodian device. No vendor with "de-identified" copies you cannot reach.

3. No third-party vendor in privileged conversations

Because nothing leaves the device, there is no third party in the Kovel sense to argue about. For a fuller treatment of the comparison, see our pieces on bot-free AI notetakers for Mac and on-device alternatives to cloud meeting AI.

Your Monday-morning checklist

1. Inventory the AI in your meetings. List every tool capable of capturing transcripts, summaries, chat, or voiceprints — including features enabled by default like Zoom AI Companion and Microsoft Copilot.
2. Map vendor retention. Pull the retention clause from Otter.ai's privacy policy, Fireflies' policy, and Zoom's privacy statement for every tool you keep.
3. Update your litigation-hold template. Add explicit categories for AI transcripts, summaries, in-meeting chat, and follow-up emails containing partial transcripts.
4. Train custodians. Disabling a bot mid-meeting does not erase what was already captured — make this explicit in training.
5. Tighten consent. Adopt the Fisher Phillips seven-step framework — obtain consent every time, from every participant.
6. Default to on-device for privileged or sensitive meetings. For board, legal, HR-investigation, and customer-negotiation calls, route capture through an on-device tool you control end-to-end.
7. Document the architecture choice. If a court later asks why you chose a particular tool, a written architecture decision citing privilege and ESI control is much stronger than "it was the default."

What about international rules?

EU-side, the same conduct that drives U.S. discovery exposure also implicates GDPR. Article 5 of the GDPR codifies data minimization and storage limitation — both directly at odds with vendors that retain meeting audio indefinitely. Article 32 requires appropriate technical and organizational measures to protect personal data, and a transcript sitting in a U.S. vendor's training pipeline is hard to defend as "appropriate." For U.S. healthcare contexts, the HHS HIPAA Privacy Rule layers BAA requirements on top of the same discovery problem.

The bottom line

The era of treating the AI bot in your meeting as a productivity feature rather than a document custodian is over. Federal courts have now ruled both ways on privilege over AI conversations in 2026; the Otter class action is moving forward in California; and Rule 37(e) gives judges the tools to punish ESI losses the moment intent crosses the threshold. The cheapest and most durable defense is architectural: if the transcript never leaves the device, opposing counsel cannot subpoena a vendor for it, employees cannot accidentally retain it for years, and your privilege analysis stops involving a third party. That is the case for on-device AI in a single sentence.

Keep your meetings off the discovery docket

Basil AI is the private, on-device AI note taker. 100% on-device transcription. No cloud. No vendor subpoena surface. No training on your conversations.