AI Use in Litigation and the Emerging Case Law

Lawyers and clients already use AI tools to summarize documents, organize facts, test arguments, and prepare for litigation. The early federal decisions addressing AI use suggest that courts are unlikely to treat AI use as inherently problematic. Instead, the emerging focus is on safeguards — what steps were taken to protect confidentiality — rather than on whether AI use categorically waives attorney-client privilege or work-product protection. While the decision in United States v. Heppner1 grabbed the attention of the legal community when it issued, that approach now appears more like an outlier than a reliable predictor of where the law is headed generally.

Heppner: AI Use as Third-Party Disclosure

The court in Heppner held that a criminal defendant waived attorney-client privilege and work-product protection by using Anthropic’s Claude to organize and analyze materials in preparation for strategy sessions with his lawyers. As to privilege, the court treated the use of the AI platform as a disclosure to a third party rather than a confidential attorney-client communication, because 1) Claude is not an attorney, 2) Heppner used it on his own initiative rather than at counsel’s direction, and 3) Anthropic’s privacy policy reserved rights to collect user data and disclose it in response to legal process. As to work-product, the court reached the same result on separate grounds: because the documents were not prepared by or at the behest of counsel, and did not reflect counsel’s litigation strategy, the doctrine did not apply.

Warner, Morgan, and Jefferies: A More Functional Approach

Three other federal decisions have generally taken a different approach to the use of AI. In Warner v. Gilbarco2, decided almost contemporaneously with Heppner, the court held that a litigant’s use of ChatGPT and other generative AI tools did not waive work-product protection. The court reasoned that work-product protection is not lost merely because material is disclosed to a third party; rather, the key question is whether the disclosure is made to an adversary or under circumstances that substantially increase the likelihood that the material will reach an adversary. Disclosure to an AI platform, the court concluded, satisfies neither condition. It described generative AI systems as “tools, not persons” and rejected broad discovery into the plaintiff’s AI use.

That same general approach appears in Morgan v. V2X3. There, the court held that a pro se litigant’s use of AI in litigation preparation could qualify for work-product protection under Rule 26(b)(3). The court drew on the reasoning of Carpenter v. United States and the Sixth Circuit’s decision in Warshak for the proposition that routing information through a third-party intermediary does not, by itself, eliminate a reasonable expectation of privacy. The court also directly addressed Heppner, distinguishing it on two grounds: Heppner was a criminal matter, and the defendant there acted entirely on his own initiative without direction from his lawyer. At the same time, however, the court did not adopt an unlimited theory of protection: it required disclosure of the identity of the AI tool used so that an assessment could be made as to whether the terms of the protective order had been complied with.

A related decision, Jefferies v. Harcros Chemicals4, reinforces the same practical concern. There, the court granted the defendants’ motion to amend the protective order to require that all discovery materials — not just those designated confidential — be processed only through closed-loop AI tools that do not use submitted data for training. The court reasoned that open AI tools make it “practically impossible” to claw back or delete information once submitted and that the risks of uncontrolled data exposure justified restricting parties to secure AI platforms.

What This Means in Practice

Taken together, these four cases do not line up neatly for or against AI use. Heppner seems to mistake AI for a person and conclude that AI use waives attorney-client privilege and work-product protections, because such use is, in fact, a disclosure to a third party. But the legal profession and bar associations have long accepted that lawyers may use cloud-based tools without waiving privilege or work-product protection so long as they exercise reasonable care. Heppner fails to meaningfully explain why AI systems should be treated differently from the broader class of cloud technologies the profession already relies on.

In contrast, Warner, Morgan, and Jefferies take a more functional approach, assuming that AI systems are no different from other cloud-based technologies and asking whether the AI platform creates a genuine risk that information will reach an adversary or escape the producing party’s control.

Looking Ahead

Over time, it seems likely that courts will settle into evaluating AI-assisted legal work under familiar confidentiality principles, asking whether reasonable safeguards were in place to prevent disclosure rather than treating AI use as a categorical problem. As appellate courts begin to address the issue, the key distinction is likely to be between AI use that creates a genuine risk of adversarial exposure or uncontrolled data access and the use of technology as part of the ordinary infrastructure of modern legal practice.

Until then, however, the emerging caselaw is just that: emerging. As a result, for lawyers and legal departments attempting to craft policies governing their use of AI, current analysis will continue to turn on the details. Platform selection matters. Contractual restrictions on storage, training use, and downstream disclosure matter. Internal policies on what may be entered into a system matter. And the degree to which the AI tool is integrated into counsel’s work may matter as well.

Questions about AI use policies or privilege considerations? Contact Nowak IP Group.