I’ve been in a lot of client calls lately where an AI note taker quietly pops into the meeting before anyone says a word. Otter.ai. Fireflies. Zoom’s built-in companion. And I get it — I use transcription tools myself. Being able to go back and verify exactly what was said is genuinely useful.
But a May 2026 New York Times DealBook piece stopped me in my tracks. Corporate lawyers are now acting as bouncers — kicking AI bots out of meetings before they start. And two federal judges issued rulings in the same month that reached opposite conclusions on whether AI tool transcripts are protected by attorney-client privilege.
If you’re an accountant or finance professional, this matters. You’re in the sensitive meetings. M&A diligence calls. Audit committee sessions. Going concern discussions. Here’s what you need to know.
From Highlight Reel to Raw Footage
Traditional meeting minutes were curated. A human note-taker applied judgment — what mattered legally, what didn’t. Offhand comments, negotiating hyperbole, quickly corrected numbers — these rarely made it into the formal record. The minutes were reviewed, approved, and framed deliberately before becoming the document of record.
AI note takers don’t work that way. They capture everything verbatim. No filter. No editorial discretion. And they store that content on third-party servers — which means when you agree to their user terms, you’re often agreeing to allow third parties to have access to your data. That’s where the legal exposure begins.
In litigation or a regulatory investigation, parties can request all documents and communications related to a topic. Every AI transcript is now a document. Lawyers who handle corporate investigations have said they fully expect regulators to start specifically requesting AI-generated transcriptions. You don’t have to be the target — a transcript from your meeting might just be relevant to someone else’s case.
The Privilege Question — And Why Courts Are Split
Attorney-client privilege protects communications between a client and their lawyer on legal matters — unless that information is shared with an outside party. That sharing breaks the privilege.
In February 2026, two federal judges ruled on nearly identical questions and reached opposite conclusions.
In the Southern District of New York, Judge Rakoff ruled that transcripts from a Claude AI session seeking legal guidance were not protected by attorney-client privilege. The reasoning: Claude’s privacy policy disclosed that data could be shared with third parties — including government regulators. No reasonable expectation of privacy, no privilege.
In Detroit, Judge Drain ruled the opposite — a pro se plaintiff’s ChatGPT transcripts didn’t have to be turned over. He rejected the argument that using ChatGPT equals disclosing to a third party.
The courts are split. The law is unsettled. And that uncertainty is the risk — because until it’s resolved, you don’t know which way a judge will rule if your transcripts end up in a dispute.
It’s Not Just Note Takers — Grammarly, Copilot, and the Rest
Here’s the part most people miss: the same third-party data processing question applies to almost any AI productivity tool you’re using.
If you’re using the free version of Grammarly, it’s almost certainly processing your text on its servers. If you’re using Copilot to summarize client financials or Notion AI to organize your notes from an audit, that content is going somewhere outside your environment. Early in the AI wave, some lawyers got into real trouble for using personal Grammarly accounts to edit legal documents. The confidentiality risk is the same for accountants.
The rule of thumb I keep coming back to: if a tool currently has an AI-enabled piece to it, it is almost certainly routing your data to a third party. Even tools with zero data retention policies are technically processing data through their servers. That might matter — and right now, no one has a definitive answer.
Two Things Accountants Should Actually Do
First: check your engagement letters. AICPA client confidentiality standards don’t have an AI exception. If your team is using AI tools on client materials — note takers in client calls, writing assistants on engagement letters, AI summarizers on financial models — you may be processing confidential data through third-party systems. That could conflict with your confidentiality clauses. Your firm needs a policy on which tools are permitted to touch client data, and your engagement letters should reflect it.
Second: answer the four governance questions. When is AI transcription permitted? Where are transcripts stored? Who can access them? And how are they handled if a litigation hold is triggered? If you’re at a large firm, these have probably been answered. If you’re not — or if you’re advising clients who haven’t thought about it — these are the questions worth asking now, before a lawsuit makes them urgent.
Key Takeaways
- AI note takers capture everything verbatim — creating a more expansive discoverable record than traditional meeting minutes
- Two federal courts issued conflicting rulings in February 2026 on whether AI tool transcripts are protected by attorney-client privilege — the law is unsettled
- The same data risk applies to productivity tools like Grammarly and Copilot when processing confidential client content
- AICPA confidentiality standards apply regardless of the tool — firms need explicit AI policies and updated engagement letter language
- Every firm should be able to answer four questions: when transcription is permitted, where it’s stored, who can access it, and how it’s handled under litigation holds
Want the CPE credit? Take the full lesson on EverydayCPE and earn 0.2 CPE credits: [lesson link]
Leave a Reply