AI in Litigation: Privilege Risks, Protective Orders, and the Path Forward

This sponsored post is authored by Aaron Marienthal and Bailey Maher, McGuireWoods

Generative AI has outpaced the legal frameworks protecting confidential and privileged information. Recent decisions show that using public AI platforms without safeguards can waive privilege, compromise work product protections, and expose sensitive materials to discovery.

Privilege and Work Product Are Not Guaranteed

In United States v. Heppner, 820 F.Supp.3d 292 (S.D.N.Y. 2026), the court held that a criminal defendant’s AI prompts and outputs were not privileged because the defendant acted without counsel’s direction and the platform’s privacy policy permitted data sharing with third parties, including government, destroying any reasonable expectation of privacy. See also Fry v. Fry, 2026 WL 2532652 (E.D. Pa. Aug. 27, 2026) (citing Heppner, refusing to seal documents uploaded to non-enterprise Claude).

Courts have since split on work product. In Warner v. Gilbarco, Inc., 820 F.Supp.3d 629, 637 (E.D. Mich. 2026), the court held that generative AI programs are “tools, not persons,” and a pro se plaintiff’s use did not constitute a disclosure to an adversary likely to waive work product protection. See also Morgan v. V2X, Inc., 2026 WL 864223, at *3–4 (D. Colo. Mar. 30, 2026) (holding pro se AI use merits work product protection, and distinguishing Heppner as criminal); but see T.B. v. Big Brothers Big Sisters of N.Y.C., 2025 WL 2443502 (N.Y. Sup. Ct. Aug. 21, 2025) (disclosing AI transcript excerpt to opposing party waived protection over entire transcript).

Where counsel directs AI use, protections are strongest. Prompts crafted by counsel are opinion work product, “virtually undiscoverable.” Tremblay v. OpenAI, Inc., 2024 WL 3748003, at *2–3 (N.D. Cal. Aug. 8, 2024); accord Concord Music Grp., Inc. v. Anthropic PBC, 2025 WL 3677935 (N.D. Cal. Dec. 18, 2025). But AI records remain discoverable when executives act independently. In Fortis Advisors LLC v. Krafton, Inc., 354 A.3d 906, 928 (Del. Ch. 2026), a CEO’s ChatGPT-generated strategy became key trial evidence.

Protective Orders: Open vs. Closed AI

Courts have begun issuing protective orders distinguishing prohibited “open” AI platforms from permissible “closed” or enterprise-grade systems. In Morgan, the court barred inputting confidential information into any AI platform unless the provider contractually prohibits training on inputs and third-party disclosure and permits deletion. 2026 WL 864223, at *7. The court acknowledged this would “bar the parties from using most, if not all, mainstream low-to-no-cost AI,” but could not “ignore the real risks associated with mainstream tools.” Id. In Jeffries v. Harcros Chems., Inc., 2026 WL 820218, at *4 (D. Kan. Mar. 25, 2026), the court extended open-system restrictions to all discovery materials, not just confidential documents. These orders also guard against opposing parties uploading produced materials into consumer AI platforms, extending protective order obligations into new territory.

The Path Forward

AI privilege law is being written in real time. The emerging consensus distinguishes unsupervised use of consumer AI insufficient to preserve privilege—from structured, counsel-directed use of enterprise tools with robust contractual safeguards. Litigants who ignore this distinction risk creating discoverable records, waiving privilege, and disclosing sensitive materials.