The lawyer received an email after the video session containing a transcript, meeting summary, and notes from the session. A useful tool to understand what was discussed, and increasingly common today. The lawyer recognized that folks should always check to see whether a meeting is being recorded, transcribed, or summarized, with the presumption that anything said may show up later. But another wrinkle had arisen. Could the session be discoverable? As courts work to keep up with technology, a wave of AI-related rulings are providing some insight into what direction the law is headed in regard to evidence and work product.
AI: not discoverable as a legal tool
In Warner v. Gilbarco, Inc., No. 2:24-cv-12333 (E.D. Mich. Feb. 10, 2026), Magistrate Judge Anthony Patti addressed a question practitioners had been debating: does using AI to draft legal materials waive work-product protection? Arguably one is taking material, sending it to a third party outside a firm like ChatGPT, and having material returned. Judge Patti found that a pro se plaintiff’s use of ChatGPT to draft legal materials did not constitute disclosure to a third party for work-product waiver purposes. AI is a tool, like a word processor or legal research platform—not a third-party advisor to whom privilege is surrendered. The fear that AI-assisted drafting could open attorney work product to discovery has chilled some from using these tools. Warner offers a useful data point, though counsel should watch for circuit-level developments—this is district court precedent, and the area is evolving quickly.
AI: discoverable as a client research tool
AI use is not always work product, particularly if a client used it to conduct research. In United States v. Heppner (Case 1:25-cr-00503-JSR USDC SDNY) a criminal defendant posed questions related to his own culpability to an AI tool. The computer he used was seized in the investigation. In this situation, since the defendant was communicating with a third party without a lawyer, there was no attorney-client privilege, nor work product, and the searches were thus discoverable. The lesson here? Caution clients about their searches. It is conceivable that in the near future we’ll be receiving production requests for our clients’ own AI searches.
The transcription problem
AI meeting transcription tools—Otter.ai, Fireflies, Microsoft Copilot’s meeting recap, and others—have become ubiquitous. The privilege analysis, however, depends entirely on who was in the “room.” A transcribed call with a client, where communications are made in confidence for the purpose of seeking legal advice, should remain privileged. The AI functions as a scribe, not a third party. But a transcribed call with a witness, a treating physician, or an adverse party’s representative? Different analysis. That transcript is likely not work product—it documents a fact-gathering exercise. Opposing counsel could argue the transcript itself is a business record, not attorney mental impressions. The work product doctrine protects thoughts about the call; the verbatim transcript of it is harder to shield. A recent Sutter Health class action suit illustrates the stakes beyond litigation. That case alleges Sutter Health used Abridge AI to transcribe confidential patient discussions and transmit them to external servers without informed consent. The healthcare context involves HIPAA, but the core problem – AI tools routing sensitive communications off-premises without adequate disclosure – is one every lawyer using free or low-cost transcription tools should reckon with.
Paying for the privilege: enterprise agreements
The two-tier AI market creates real risk for practitioners who don’t read the fine print. Enterprise agreements and Business Associate Agreements (BAAs) typically include data isolation commitments. These commitments typically include parameters where the information put in doesn’t train the model, doesn’t get shared, and is subject to meaningful contractual protections. Consumer and freemium products often operate under different terms. That free transcription app? The terms of service may allow the provider to use recordings to improve their model. Bar rules on confidentiality don’t have a carve-out for convenient tools with fuzzy data practices. Before deploying any AI tool that touches client communications consider the risks associated with freemium versions and confirm what the provider does with the data. Enterprise-tier products with explicit confidentiality protections are appropriate for client matters.Â
Outro
Back to the lawyer and the post-meeting transcript. Because the lawyer worked with a good information technology team, the lawyer had stayed current on technology and had adopted a business associate agreement for the transcription tool. The conversation had been with the client, with the client’s approval. The meeting the lawyer had just had was not the type that would make the transcription discoverable. At least not with the current state of the law. With the direction the law was trending the lawyer was comfortable continuing this practice. But the intersection between law and AI was fraught with frontier concepts, and one where practitioners needed to stay current to avoid running into issues.
Bio:
Miles B. Cooper is a partner at Coopers LLP, where they help the seriously injured, people grieving the loss of loved ones, preventable disaster victims, and all bicyclists. Miles also consults on trial matters and associates in as trial counsel. He has served as lead counsel, co-counsel, second seat, and schlepper over his career, and is an American Board of Trial Advocates member.
