Artificial intelligence is rapidly transforming litigation practice—reshaping how cases are investigated, analyzed and litigated. While there is no doubt that these tools can offer significant efficiencies, they also introduce new legal, ethical and regulatory risks that courts are beginning to scrutinize very closely.
As more parties use generative AI in litigation, courts are being asked a straightforward question: if someone uses AI to help think through claims, test arguments or draft filings, can the other side demand those materials in discovery? So far, courts are not treating AI as a category of its own. They are applying familiar rules instead: looking at why the material was created, whether it reflects legal strategy, whether counsel directed the work, whether the party later relied on it and whether confidentiality was preserved.
Recent rulings from New York, Michigan and California show that the answer remains highly fact-dependent.
In United States v. Heppner,1 Judge Jed S. Rakoff of the Southern District of New York held that AI-generated materials were not protected from disclosure. A criminal defendant used an LLM to generate materials relating to possible defenses after receiving a grand jury subpoena. The court found no attorney-client privilege because the LLM was not a lawyer and the platform’s privacy terms defeated any reasonable expectation of confidentiality. The court also found no work-product protection because counsel had not directed the work and the materials did not reflect counsel’s strategy.
Warner v. Gilbarco, Inc.2 ruled the other way. Warner is a pro se employment-discrimination action in which defendants sought “all documents and information” concerning the plaintiff’s use of third-party AI tools in connection with the lawsuit. Magistrate Judge Anthony P. Patti of the Eastern Michigan District Court denied the request. Although the court also found the motion untimely, it went further and held that the requested AI-related materials were not discoverable because they were not relevant or, at most, only marginally relevant and disproportionate. The court also held that, even if discoverable, the materials were protected work product and it rejected the waiver argument on the ground that “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.” Just as importantly, the court noted there was no evidence that the plaintiff had uploaded confidential material in violation of a protective order.
A decision out of California falls somewhere in between. In Tremblay v. OpenAI,3 a copyright case in the Northern District of California, the plaintiffs used ChatGPT before filing suit to test whether it could summarize their books. The court held that prompts crafted by counsel could reflect counsel’s mental impressions and therefore qualify for strong work-product protection, while limiting any waiver to material actually relied on the complaint.
Taken together, these cases suggest a practical point more than a sweeping legal rule. Courts are not saying that anything involving AI is automatically discoverable. However, they are also not creating a safe harbor for AI use in litigation. As with other sensitive materials, the details matter.
Clients should assume, for now, that prompts, outputs, uploaded documents and saved chats may later become relevant in discovery. If generative AI is going to be used in connection with a dispute, it is best done thoughtfully: under counsel’s direction, on a platform that has been vetted for confidentiality and with clear internal guidance about what can and cannot be entered into the system. Businesses should also make sure their document preservation and AI governance practices account for litigation-related AI use.
Haynes Boone will continue to monitor developments in this area and provide updates as the law evolves.
1 United States v. Heppner, No. 25 CR. 503 (JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026). Judge Rakoff’s decision from the bench was issued on February 10, 2026. See Transcript of Record at 2:25-3:15, U.S. v. Heppner, No. 25 CR. 503 (JSR) (S.D.N.Y. Feb. 10, 2026).
2 Warner v. Gilbarco, Inc., No. 2:24-CV-12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026).
3 Tremblay v. OpenAI, Inc., No. 23-cv-03223-AMO, 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024).
