Attorney Client Privilege Meets AI, Part 2

Here is my recent Daily Record column. My past Daily Record articles can be accessed here.

****

Attorney Client Privilege Meets AI, Part 2

It’s time to once again revisit artificial intelligence and privilege. After writing about these issues in last week’s column, there’s another case to add to the mix: 

Last week, I covered a recent S.D.N.Y. Order from District Court Judge Jed S. Rakoff in U.S. v. Heppner. Judge Rakoff concluded that when a defendant used the free version of Claude’s generative artificial intelligence (AI) tool to draft defense strategy materials that he later handed over to his attorney, that privilege did not apply. 

He reasoned that privilege requires the existence of a “trusting human relationship,” which does not exist “between an AI user and a platform such as Claude.” Therefore, sharing information with Claude would waive any applicable privilege, since it’s no different than sharing it with any other third party. In reaching his conclusion, he suggested that AI tools are inherently different from other technologies long relied upon in the legal profession, including email and cloud computing. 

I disagreed with that rationale, suggesting that Judge Rakoff’s approach would not withstand the test of time and that generative AI should be treated not as a third party to whom information is disclosed. Instead, it’s more akin to a tool that processes privileged communications.

On the same day that Judge Rakoff weighed in on AI and privilege, another court reached an inapposite conclusion. In Warner v. Gilbarco, Inc., No. 2:2024cv12333 - Document 94 (E.D. Mich. 2026), the underlying facts and issues addressed were different from those in Heppner, but the ultimate conclusion was nevertheless instructive. 

In Warner, the plaintiff appeared pro se in an employment discrimination matter. United States Magistrate Judge Andrew P. Patti’s Order addressed the parties' discovery disputes. One area of contention was the defendant’s requests relating to the plaintiff’s use of generative AI tools in connection with the lawsuit. The Court declined to grant the request for “all documents and information concerning her use of third-party AI tools in connection with this lawsuit, as requested in Defendants’ discovery requests,” since they were used in preparation for litigation. The Court chastised the defendant’s single-minded focus on the AI tools, stating that the “preoccupation with Plaintiff’s use of AI needs to abate.”

Next, the Court explained that the plaintiff did not “waive work product-protection by using ChatGPT” since she did not disclose “work-product to an adversary or in a way likely to get in an adversary’s hand.” The Court reasoned that ChatGPT is a tool, not a person, even if it may “have administrators in the background.”  Concluding otherwise would “nullify work-product protection in nearly every modern drafting environment…” especially since “no cited case orders the production of what Defendants seek here: a litigant’s internal mental impressions reformatted through software.”

In other words, the Warner court recognized a reality that Heppner missed: the output of an AI tool isn't a third-party statement. Instead, it is the litigant’s own 'mental impressions reformatted through software.' If the software is merely the conduit for the thought, it shouldn't dictate the protection.

Importantly, the reasonable expectation of confidentiality does not exist in a vacuum. If a platform’s terms allow user inputs to be stored, reviewed, or used beyond the immediate interaction, courts may view that differently than a system with clear contractual restrictions on data use. The technology label alone is not dispositive, but the data practices may be.

Despite two very different fact patterns and approaches, the issue ultimately boils down to this question: Should generative AI be treated as a third party or a tool that processes attorney work product?

How courts resolve that question will impact more than just generative AI use. It will influence how and whether lawyers can continue to use other long-accepted technologies, including email and cloud platforms, to manage confidential information in everyday practice.

It is well established that these types of technology providers are simply service vendors or agents that facilitate legal representation. Entering confidential information into these tools should not, and does not, impact privilege or waive work product. To hold otherwise would call into question a wide range of everyday tools used in law firms, significantly disrupting law firm workflows and court dockets across the country.

Ultimately, both attorney-client privilege and work product protection are designed to preserve the confidentiality necessary for effective representation. That objective is not served by treating widely used drafting technology as an adversarial third party. The focus should remain on whether confidentiality was reasonably maintained, not on whether the tool involved is labeled “AI.” 

Nicole Black is a Rochester, New York attorney, author, journalist, and Principal Legal Insight Strategist at 8am, the team behind MyCase, LawPay, CasePeer, and DocketWise.She is the nationally-recognized author of "Cloud Computing for Lawyers" (2012) and co-authors "Social Media for Lawyers: The Next Frontier" (2010), both published by the American Bar Association. She also co-authors "Criminal Law in New York," a Thomson Reuters treatise. She writes regular columns for Above the Law, ABA Journal, and The Daily Record, has authored hundreds of articles for other publications, and regularly speaks at conferences regarding the intersection of law and emerging technologies. She is an ABA Legal Rebel, and is listed on the Fastcase 50 and ABA LTRC Women in Legal Tech. She can be contacted at niki.black@mycase.com.