Kang Haggerty News
The recent wave of litigation involving generative AI in discovery is clarifying when prompts, responses, and chat histories are protected, revealing that courts distinguish between attorney-directed use and independent client research, between factual material and litigation strategy, and between the confidentiality protections of privilege and the broader shield of work product. While consumer AI platforms may defeat confidentiality under existing privacy policies, work-product protection can survive AI assistance if the materials were created for litigation purposes and not disclosed to adversaries.
In the November 5, 2026 edition of The Legal Intelligencer, Kelly Lavelle writes, “The Evolving Law of Generative AI in Discovery.”
The use of generative AI in litigation is testing the application of established discovery rules to prompts, responses, and account histories that may contain both factual information and litigation strategy. Recent decisions examine whether a client’s independent use of AI defeats confidentiality, which portions of an expert’s AI-assisted research and drafting must be disclosed, and whether litigation-preparation materials can be obtained by subpoenaing the provider. These decisions distinguish between attorney-client privilege and work-product protection while underscoring the importance of counsel’s involvement, the platform’s terms, and the purpose for which the materials were created.
The widely discussed decision in United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026) illustrates how a client’s independent use of generative AI can undermine a claim of privilege. In rejecting that claim, the court focused on the absence of confidentiality in the defendant’s exchanges with Claude. Judge Jed Rakoff, confronting what he called a question of first impression nationwide, held that a represented defendant’s self-directed exchanges with the consumer version of Claude were neither attorney-client privileged nor work product. The court emphasized that Claude is not a lawyer, that counsel had not directed the defendant’s use, and that the applicable privacy policy permitted the collection, training use, and disclosure of user information to third parties, including regulators. Those terms defeated any reasonable expectation of confidentiality. Sending the outputs to counsel afterward did not make them privileged. The court separately rejected work-product protection because the materials were not prepared at counsel’s direction and did not reflect counsel’s strategy. The court also suggested that the result could differ if counsel had directed the defendant’s use of AI, leaving open whether those exchanges could be protected by the attorney-client privilege.
The confidentiality concerns addressed in Heppner also arose in Fry v. Fry, 2026 WL 2531956 (E.D. Pa. Aug. 27, 2026). There, a pro se petitioner disclosed that he had uploaded redacted documents to nonenterprise versions of Claude and Claude Cowork. In declining to seal those documents in their entirety, Judge Karen Marston considered the disclosure to Anthropic and the platform’s privacy and training policies, citing Heppner. Although Fry did not decide whether AI prompts or outputs were privileged or discoverable, it illustrates that uploading litigation materials to a consumer AI platform can affect a court’s assessment of confidentiality.
Apart from questions of confidentiality and privilege, courts are recognizing that AI prompts and responses constitute electronically stored information subject to preservation obligations. In Miller v. Regions Bank, 833 F. Supp. 3d 1201 (N.D. Ala. 2026), the court expressly identified an attorney’s ChatGPT query history as ESI and found that, at a minimum, counsel had a clear duty to preserve it once the court ordered its production. When counsel instead deleted the account three days later and requested a prorated refund to terminate access immediately, the court found bad-faith spoliation and drew an adverse inference. Although the court questioned whether Rule 37(e) governed the destruction of information sought for its own sanctions inquiry, it concluded that its finding of bad faith supported an adverse inference under either that rule or its inherent authority. The decision confirms that ChatGPT history is ESI and must be considered when identifying potentially discoverable information and determining what must be preserved.
The separate question of whether an AI chat log must be produced in full, or reviewed for protected material, was addressed in Villanueva v. Las Vegas Metropolitan Police Department, 2026 WL 2942348 (D. Nev. Sept. 29, 2026). There, a plaintiff’s expert disclosed at his deposition that he had used ChatGPT to research and draft his report but had not disclosed that use in response to a subpoena for “all data, documents, and other information” he considered in forming his opinions. Reviewing a magistrate’s order compelling the production of the expert’s entire ChatGPT history log, the district court distinguished among its contents.
The court found that AI-generated case summaries the expert consulted were “facts or data considered” under Rule 26(a)(2)(B)(ii) and had to be produced. The summaries were subject to disclosure because the expert considered them when forming his opinions, even if he ultimately did not rely on them. The court also noted that the log was the only place the defense could examine allegedly hallucinated case citations. But the expert’s ChatGPT-generated deposition outline and questions, prompts and draft report material were protected under the trial-preparation protections of Rule 26(b)(3) and the protection for draft expert reports “regardless of the form in which the draft is recorded” under Rule 26(b)(4)(B). Because ChatGPT “is not a person” but a research, drafting, and editing application, the expert’s use did not waive those protections. The decision therefore required a distinction between discoverable source material and protected research and drafting activity within the same AI history.
That reasoning echoed Warner v. Gilbarco, 820 F. Supp. 3d 629 (E.D. Mich. 2026), where the court also refused to compel a pro se litigant’s production of all documents and information concerning her use of third-party AI tools in connection with the lawsuit. The court recognized the self-represented litigant’s right to assert work-product protection and held that using an AI platform did not itself waive that protection. Waiver requires disclosure to an adversary or disclosure in a manner likely to place the material in an adversary’s hands. The court explained that ChatGPT and other generative AI programs) are “tools, not persons,” even if they may have administrators somewhere in the background.
That protection may also apply when discovery is directed to the provider rather than the litigant. In Assini v. Hayward, 261 N.Y.S.3d 915 (N.Y. Sup. Ct. 2026), plaintiffs subpoenaed OpenAI for a pro se defendant’s prompts, uploads, and outputs relating to the litigation. The court quashed the subpoenas, applying New York’s protection for litigation-preparation materials and finding persuasive federal authority protecting a pro se litigant’s AI-assisted preparation. The decision demonstrates that seeking the material directly from the platform does not eliminate the need to consider whether its contents are protected.
Taken together, these decisions demonstrate that protection for AI prompts and responses depends on the circumstances of their use. Villanueva and Warner appear to take a different approach from Heppner, although part of that difference concerns the protection asserted. Neither Villanueva nor Warner treated use of the platform as automatically waiving work-product protection, while Heppner made the applicable privacy terms decisive for confidentiality. The distinction between privilege and work-product waiver helps explain the results, but does not fully resolve the courts’ differing approaches to work product, particularly the significance of counsel’s involvement.
The decisions thus require courts to distinguish among factual source material, independent client research, and protected litigation preparation, all of which may appear within a single chat history. Neither the subject of the exchange nor the use of AI alone determines whether disclosure is required. Where courts draw those lines will determine how much of that history becomes available to an opposing party.
Kelly A. Lavelle is senior counsel at Kang Haggerty. She focuses on e-discovery and information management, from preservation and collection to review and production of large volumes of electronically stored information. Contact her at klavelle@kanghaggerty.com.
Reprinted with permission from the November 5, 2026 edition of “The Legal Intelligencer” © 2026 ALM Global, LLC. All rights reserved. Further duplication without permission is prohibited. Request academic re-use from www.copyright.com. All other uses, submit a request to asset-and-logo-licensing@alm.com. For more information visit Asset & Logo Licensing.