Client Use of Generative AI During Litigation May Make Privileged Information Discoverable
08/11/26
Client Use of Generative AI During Litigation May Make Privileged Information Discoverable
Recent decisions addressing client-litigants’ use of large language model AI tools reveal an emerging, fact-sensitive split over whether AI prompts, outputs, and related communications are protected from discovery. The split illustrates the practical risk for business clients already in litigation or arbitration: independent use of consumer AI tools may turn private research, timeline-building, discovery drafting, or strategy-testing into discoverable material rather than protected attorney-client communication or work product. That risk increases when a client inputs confidential facts, counsel’s advice, legal theories, witness information, or settlement considerations into tools whose terms and data practices may not preserve confidentiality.
Recent Cases Highlight a Developing Split
Two recent district court decisions show why clients should not assume AI-related communications will remain protected.
In United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026), Judge Jed S. Rakoff held that a litigant’s exchanges with Claude were not protected by attorney-client privilege or the work-product doctrine. The defendant used Claude on his own, shared information learned from counsel and later gave the AI-generated materials to his lawyers. The Court found no privilege because the communications were not with counsel and were not reasonably confidential. Judge Rakoff also rejected work-product protection because counsel had not requested, prepared, or supervised the materials. In support of his decision, Judge Rakoff stressed that the litigant “could have had no ‘reasonable expectation of confidentiality in his communications’ with Claude” and that by sharing his notes with Claude, he had shared them with the “equivalent … of a third-party.”
Heppner warns that client-generated AI materials may not be privileged and do not become privileged simply because the client later sends them to counsel or hopes they will assist the defense. (For more information on Heppner, see our March 5, 2026 Legal Update.)
By contrast, Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), reached a different result. Magistrate Judge Anthony P. Patti denied discovery into a pro se plaintiff’s use of ChatGPT and similar tools while preparing her case. Because the materials were prepared in anticipation of litigation by a party representing herself, the court treated them as protected work product and declined to find a waiver of that protection from AI use alone.
Warner does not create a blanket protection for AIbel use. Read with Heppner, it shows that courts will likely focus on the platform used, the information disclosed, whether counsel directed the work, and whether the materials reveal legal strategy or independent client research.
Discovery Requests Targeting AI Use Are Coming
Discovery over AI use is likely to become more common. Opposing parties may seek prompts, outputs, and related information about whether AI was used to draft discovery responses, organize facts, evaluate claims, assess settlement, or prepare witnesses. Those requests may become part of ordinary electronically stored information discovery.
Litigants therefore should not assume traditional privilege objections will defeat discovery into client AI use. If a client has disclosed sensitive facts, legal theories, or counsel’s advice to a consumer AI tool, there is risk that those actions will not be considered privileged or may even waive privilege. And even if not, the resulting fights over privilege and work-product protection will likely add cost and risk to the litigation.
Practical Takeaways for Clients
Clients should be extraordinarily cautious about inputting confidential facts, attorney advice, legal theories, or strategy into consumer AI tools without counsel’s involvement. Also, litigants should consider a document-management and communication structure that preserves confidentiality and protects attorney-client privilege and work-product claims.
Please contact the author, John T. Below, or any member of Bodman’s Artificial Intelligence Team, if you have questions regarding any of the information above. Bodman cannot respond to your questions or receive information from you without establishing an attorney-client relationship and clearing potential conflicts with other clients. Thank you for your patience and understanding.