AI On Trial: Morgan V. V2X Draws New Lines On Work Product Protection And Confidentiality

As artificial intelligence tools become ubiquitous in litigation, federal courts are increasingly confronting novel questions about how longstanding doctrines of work-product protection and confidentiality apply when parties route litigation materials through public AI platforms. In the span of just seven weeks, three federal courts have issued notable decisions at this intersection: Warner v. Gilbarco, Inc. (E.D. Mich. Feb. 10, 2026), United States v. Heppner (S.D.N.Y. Feb. 17, 2026), and Morgan v. V2X, Inc. (D. Colo. Mar. 30, 2026). Together, these decisions begin to define an emerging (yet still unsettled) framework governing AI use in discovery, with significant practical consequences for litigants, attorneys, and in-house legal teams.

In the most recent decision, Morgan v. V2X, Magistrate Judge Maritza Dominguez Braswell addressed a dispute in an employment discrimination case between a pro se plaintiff and a corporate defendant over whether the plaintiff’s choice of AI tool warranted work-product protection and whether the court should amend the protective order to restrict AI use in connection with confidential information. Plaintiff Archie Morgan argued that disclosing the identity of his AI tool would reveal his mental impressions and litigation strategy, invoking Federal Rule of Civil Procedure 26(b)(3). The court agreed that work-product protections apply to a pro se litigant’s use of AI, emphasizing that Rule 26(b)(3) broadly protects materials that any party—not merely counsel—prepares in anticipation of litigation, and that conditioning protection on the involvement of an attorney “finds no support in the rule’s text and would further disadvantage unrepresented litigants.” However, the court held that the plaintiff had not met his burden of showing that merely disclosing the name of his AI tool would reveal protected mental impressions, and it ordered disclosure of the tool’s identity so that the defendant could assess whether confidential information had been compromised. On the protective order question, the court rejected both parties’ competing proposals and instead crafted its own AI-specific provision, which prohibits the input of confidential information into any AI platform unless the provider is contractually prohibited from using inputs for model training or disclosing them to third parties, and must afford users the ability to delete all confidential data on request.

The Morgan decision sits at a critical juncture between two earlier rulings. In Warner v. Gilbarco, the court similarly found that a pro se litigant’s use of ChatGPT and other generative AI platforms warranted work-product protection, reasoning that AI tools are “tools, not persons,” and that disclosing materials to them does not constitute disclosure to an adversary or create a substantial risk that the materials will reach an adversary. By contrast, in United States v. Heppner, the Southern District of New York declined to extend work-product protection to a represented criminal defendant who independently used AI without direction from his attorney, reasoning that work-product protections have historically shielded the mental impressions and legal theories of attorneys.

In Morgan, Judge Braswell acknowledged Heppner but distinguished it on two grounds: first, that the criminal context in Heppner is governed by a different legal framework than the Federal Rules of Civil Procedure; and second, that the gap between party and counsel that existed in Heppner does not exist in the pro se context, where the litigant holds the dual role of “party and advocate.” The common thread across all three decisions is a recognition that courts cannot analyze AI use in litigation through a binary lens—courts are willing to extend traditional protections to AI-generated work product, but the scope and strength of those protections depend heavily on context, including the litigant’s representation status, the nature of the proceeding, and whether counsel’s work connected to the AI use.

For practitioners and in-house teams, the practical implications are significant:

First, the Morgan court’s ruling that simply using a public AI tool does not automatically waive work-product protection is a welcome development, particularly for pro se litigants and resource-constrained parties. But the court was equally clear that this protection has limits: parties must be prepared to justify claims of privilege with specificity, and conclusory assertions that tool selection reveals litigation strategy will not suffice.

Second, the court’s AI-specific protective order provision effectively bars the use of most mainstream, consumer-grade AI tools—such as standard ChatGPT, Claude, or Gemini—to process confidential discovery materials, unless the provider offers contractual guarantees against data retention, model training, and third-party disclosure. As the court itself acknowledged, only enterprise-tier AI platforms will likely meet this standard, creating what the court called “a growing problem in the age of AI” for litigants who cannot afford those tools.

Third, courts appear increasingly inclined to require transparency about AI use in connection with protected materials, even where work-product protections apply—suggesting that, at a minimum, parties may not shield the identity of AI tools from disclosure.

Takeaways

The Morgan decision, viewed against the backdrop of Warner and Heppner, signals thatAI use in litigation is now a discoverable, contestable dimension of case management with real evidentiary and procedural consequences. Courts have shown they will not defer to the parties on these issues and may impose obligations the parties never anticipated. Practitioners should take note of the following:

• AI governance can no longer be siloed as an IT or compliance function.It must be integrated into litigation strategy from the outset, including during meet-and-confers, protective order negotiations, and discovery planning.
• The doctrinal landscape is fracturing along fault lines largely unrelated to the technology itself.Whether work-product protection applies turns on factors like representation status, the civil or criminal nature of the proceeding, and the degree of attorney involvement.
• Institutional litigants should note that the Morgan court’s protective order standard will likely disadvantage pro se parties.Aggressive protective order proposals may invite judicial scrutiny of a party’s own AI practices.

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.