We recently examined two court orders addressing a party’s use of generative artificial intelligence (GenAI) tools and how that use may impact whether the prompts and outputs of those tools are entitled to privilege protection.

In February 2026, in United States v. Heppner (Heppner)1, the US District Court for the Southern District of New York held that documents generated by a defendant through the publicly-available GenAI tool Claude were not protected by attorney‑client privilege or the work-product doctrine and could be produced. That order garnered significant attention from legal practitioners and companies alike. Our initial article discussed Heppner and highlighted that subsequent AI-privilege decisions by other courts may differ.

This article continues our examination of the Heppner case and highlights two other recent federal court decisions that resolved similar GenAI-privilege issues: Warner v. Gilbarco, Inc., (Warner)2 and Archie Morgan v. V2X (Morgan).3 Both cases provide additional guidance on how courts may treat potentially privileged and confidential information processed by GenAI tools—particularly publicly-available tools—in both criminal and civil contexts.

In this article, we briefly summarize these recent decisions, outline the courts’ findings in Warner and Morgan, identify where these cases diverge and align with Heppner, and provide key takeaways for companies looking to safeguard privilege. We also address the Morgan court’s rulings regarding the use of protective orders, and the role protective orders can serve to provide guardrails for the parties’ input of confidential information into GenAI tools.

Summary of decisions addressed

  • United States v. Heppner: In a federal criminal proceeding, the US District Court for the Southern District of New York held that neither the attorney-client privilege nor the work product doctrine applied to GenAI prompts and outputs, where defendant used a publicly available version of Claude to prepare for a meeting with his attorney in anticipation of a potential indictment.
  • Warner v. Gilbarco, Inc.: In a federal civil proceeding, the US District Court for the Eastern District of Michigan held that the work product doctrine applied to GenAI prompts and outputs stemming from a pro se plaintiff’s use of a publicly available GenAI tool under Fed. R. Civ. P. 26(b)(3).
  • Archie Morgan v. V2X, Inc.: In a federal civil proceeding, the US District Court for the District of Colorado held that the work product doctrine applied to GenAI prompts and outputs stemming from a pro se plaintiff’s use of a publicly available GenAI tool under Fed. R. Civ. P. 26(b)(3). The court also amended the parties’ stipulated Protective Order to prohibit the parties from inputting confidential information into GenAI tools that may lack appropriate confidentiality and security protections.

Courts find work-product protection for pro se litigant’s use of public GenAI tool

When the US District Court for the Eastern District of Michigan and the US District Court for the District of Colorado were confronted with the question of whether work product privilege applied to the use of GenAI, both courts reached a different conclusion from the Heppner court.4 In both cases, the court relied on Rule 26(b)(3) of the Federal Rules of Civil Procedure to find that a pro se plaintiff’s GenAI-assisted materials were protected as opinion work product. Whereas Warner was issued prior to Heppner, Morgan specifically distinguishes from Heppner and cites favorably to the Warner decision.

Warner v. Gilbarco

In Warner, an employment discrimination case, the court found that the work product doctrine applied to a pro se litigant’s use of a publicly-available version of a GenAI tool to prepare an internal legal analysis.

Specifically, the Warner pro se plaintiff relied heavily on a publicly available version of a GenAI tool to assist in drafting legal documents and preparing legal analysis to support her claims. The defendants moved to compel her to produce all documents and information related to her use of GenAI tools, arguing that her use of the GenAI tool constituted disclosure to a third party (the GenAI platform owner), and thus that the use waived work-product protection.

In response, the plaintiff argued the GenAI-related materials reflected her “internal analysis and mental impressions,” and were protected from disclosure as opinion work product. The court agreed, finding that the materials were protected. The court additionally found that GenAI tools are “tools, not persons” for the purpose of a third-party waiver analysis.5

Morgan v. V2X

In Morgan, another employment discrimination and retaliation case, the court addressed how work-product protections and confidentiality obligations apply when a pro se litigant uses AI tools in civil litigation. At issue was a discovery dispute concerning the use of GenAI and whether the parties’ stipulated Protective Order should be amended to regulate GenAI use.~~~~

Morgan’s former employer, V2X, filed a motion compelling Morgan to disclose the name/identity of any GenAI tools he used in connection with V2X’s confidential information, and to amend the existing stipulated protective order to account for the use of GenAI tools in litigation.

The court determined that this dispute raised two questions:

  1. Whether work product protections apply to a pro se Plaintiff's use of AI, and in particular, to his selection of a particular AI tool?
  2. To what extent should a protective order restrict AI use, particularly with respect to the disclosure of confidential materials produced in discovery?

On the first issue, Morgan, proceeding pro se, opposed disclosure of the identity of his GenAI tool, arguing that both his GenAI tool selection and his interactions with the GenAI tool, were protected under the work product doctrine as defined in Federal Rule of Civil Procedure 26(b)(3). The Morgan court initially focused on whether the work product doctrine applied to protect a non-attorneys’ work product where the party is proceeding pro se, concluding that a pro se plaintiff’s work product should be treated with the same protection as work product prepared by a represented plaintiff’s attorney. The Morgan court acknowledged that the Heppner decision may appear contrary to this finding, but distinguished Heppner on the basis that first, Heppner was a criminal proceeding and Rule 26(b)(3) broadly protects work product of a party and not merely their counsel, and second, that unlike the Heppner defendant who had retained counsel and was purportedly acting “entirely apart from his lawyer” and not at his lawyer’s direction, there was no need in this case to examine any potential gap between whether the client was acting at the direction of their attorney in their use of a GenAI tool or not because a pro se party acts as both the client and attorney.

While the court does not further examine Heppner, the court also cites to Warner to support its finding that privilege was not waived simply through the use of a publicly available AI tool, because while work product generated using the GenAI tool may have been available to the AI tool developer, it was not disclosed to any adversarial party. The court states:

[G]iven how AI tools function, it is entirely reasonable for a person to expect some privacy and confidentiality when interacting with these tools, even though they understand a third party is behind the tool collecting and storing their information.

In any event, work product protections are typically waived by disclosure to an adversary, or in circumstances that substantially increase the likelihood that an adversary will obtain the materials.6

This reasoning appears to further depart from the Heppner court’s conclusion that the use of a publicly available AI tool automatically eliminates a user’s expectation of privacy in their use of the tool.

While Morgan concluded that the plaintiff could assert work product protection with respect to AI tool inputs and outputs, the court declined to extend such work product protections to the name of the tool itself because plaintiff Morgan failed to show how disclosing the name or identity of the GenAI tool he used would reveal his mental impressions or strategic thinking. Because V2X had a legitimate need to assess whether its confidential information was compromised through its disclosure into an AI tool, the court ordered Morgan to disclose the name of any GenAI platform to the extent it was used to process V2X’s confidential information.

Court addresses use of public GenAI tools in context of protective orders

In Morgan, because the parties’ original protective order broadly prohibited disclosure to third parties, the court also examined whether and how to modify the protective order to govern the disclosure of confidential information into a GenAI platform. Both parties were aligned on the need for a clarifying amendment regarding restrictions on GenAI use; however, they diverged on the specific language to be implemented. In particular, Morgan argued that restricting GenAI use would unfairly disadvantage a pro se litigant, compared to a corporate defendant with significant technological resources. Each party submitted their own proposed supplemental language to the court. The court rejected both submissions and amended the protective order as follows:

No party or authorized recipient may input, upload, or submit CONFIDENTIAL Information into any modern artificial intelligence platform, including any generative, analytical, or large language model-based tool (“AI”), unless the AI provider is contractually prohibited from: (1) storing or using inputs to train or improve its model; and (2) disclosing inputs to any third party except where such disclosure is essential to facilitating delivery of the service. Where disclosure to a third party is essential to service delivery, any such third party shall be bound by obligations no less protective than those required by this Order. In addition, the AI provider must contractually afford the party or authorized recipient the ability to remove or delete all CONFIDENTIAL information upon request. A party intending to use AI that it contends meets these requirements must retain written documentation of these contractual protections.7

The court appears to have taken into account defendant’s concerns that, depending on the GenAI provider’s terms of service, defendant’s confidential information could be used to train the GenAI provider’s GenAI model, opening the door for a third party to receive and use the parties’ confidential information in violation of the protective order. At the same time, the Morgan court expressly states that the protective order should not leave plaintiff without the benefits of GenAI, and that GenAI can still be used in ways that do not involve uploading defendant’s confidential information. While acknowledging that this restriction likely disadvantages pro se litigants, the court concluded that, on balance, the risk of compromising confidential information justified this limitation. Of note, this acknowledgement that information uploaded into a GenAI platform could be used (and thus disclosed to) a third party seems to align with the Heppner court’s underlying analysis regarding confidentiality.

Key takeaways on recent GenAI decisions and privilege protections

As more courts weigh in on the applicability of privilege in the GenAI context, we expect this to continue to be an area of dispute. In particular, these recent cases all address what seems to be the same subject—the use of publicly available GenAI tools—and yet appear to diverge on how they examine privilege applicability and waiver, particularly for materials that may be covered by the work product doctrine. Whereas Heppner appears to take a more bright line approach that neither attorney client privilege nor work product doctrine protections could exist due to the lack of confidentiality inherent to the use of a publicly available GenAI tool, the Warner and Morgan courts both conclude that work product protection could exist and instead examine whether the work product  privilege was waived based on the facts around the party’s specific use of a publicly available GenAI tool.

While the parties in both Warner and Morgan were pro se plaintiffs, that is not the sole factor distinguishing these decisions from Heppner. Importantly, Federal Rule of Civil Procedure 26(b)(3) not only protects work product materials prepared by or at the direction of an attorney, but also materials prepared “by another party or its representative.” This distinction may be particularly relevant to the extent criminal and state rules define the work product doctrine more narrowly. Because pro se plaintiffs act as both the client and attorney, this may make it easier to demonstrate that GenAI materials prepared by a pro se plaintiff qualify for protection under Rule 26(b)(3), but courts relying on Warner and Morgan could reach a similar conclusion outside of the pro se context assuming there is sufficient evidence that the parties’ use of GenAI related to work product materials prepared in anticipation of litigation under Rule 26(b)(3).

Both Warner and Morgan decisions also depart from Heppner by noting that GenAI is a “tool,” not a person that would raise potential third party waiver considerations. Relatedly, the Morgan court appears to take a different position than the Heppner court regarding whether the use of a publicly available GenAI tool by itself eliminates any expectation of privacy the user has in the tool such that privilege protections are automatically waived. The Morgan court raises an analogue in an individual’s use of email, citing prior circuit decisions holding that the mere fact that information is held by a third-party intermediary, like an internet service provider, does not extinguish a user’s expectation of privacy and confidentiality. While the Warner and Morgan courts only address the applicability of the work product doctrine and do not address attorney-client communications, these decisions could be used to support the argument that the use of a publicly available AI tool does not necessarily automatically waive privilege protections where it is used as a tool to create attorney-client communications (for example, a client using GenAI to draft an email to their attorney).

The Morgan court further notes that when examining waiver under the work product doctrine, courts typically waive protection where this information is disclosed to an adversary. Here, the court determined it was “highly unlikely” that GenAI interactions with a third party AI tool developer would fall into the hands of the opposing party (who was not the developer of the AI tool in question). Indeed, when the Heppner court reached a different conclusion, it relied upon In re: OpenAI Inc. Copyright Infringement Litigation,8 which opined on users’ expectation of privacy in prompts submitted to a GenAI tool owned by OpenAI, where OpenAI itself was the adversarial party in the dispute. It is critical to note that the OpenAI court did not address these issues in the context of privilege, and did not opine on whether users may have an expectation of privacy in their prompts and outputs with respect to a non-AI developer third party. With this context in mind, Morgan and Warner suggest that even where a publicly available GenAI tool is used to generate potential privileged materials, particularly under the work product doctrine, privilege may apply.

What is clear from these recent decisions is that, subject to the right conditions, privileged materials created in connection with the use of an AI tool, even a publicly available AI tool, may be protected. Considerations around whether the tool is publicly available, whether counsel has directed the client to use the tool and whether the purportedly privileged material is made available to an adversary, among other things, will continue to be relevant in examining privilege protection and waiver.

While we have already addressed the key issues companies should consider around GenAI tools and downstream risks to privilege, we anticipate that companies looking to safeguard privilege will likely want to instruct employees to use enterprise (not publicly available) GenAI tools in their legal matters, and guide employees on the appropriate use of these tools particularly when they are being used to support legal investigations.

Practical considerations on recent GenAI decisions and the role of protective orders

As GenAI use becomes more prevalent, we can expect parties may use GenAI tools to assess materials they receive in litigation, including confidential information produced pursuant to a protective order. Given the unique risks that GenAI tools pose, companies should likely examine how they can strengthen protective orders to address these novel risks.

Consider whether existing and future protective orders contemplate the parties’ use of GenAI

This goes far beyond the GenAI-directed document review. In the document review context, we have seen parties take a similar approach to GenAI-directed review as they did with Technology Assisted Review (TAR) to outline requirements to notify the parties where TAR will be used to cull documents from review, or establish a review and validation framework for the parties’ use of TAR. These requirements are often described in an ESI Protocol. We expect that parties should consider and likely are already considering applying similar requirements to document review in the GenAI context.

Parties should also address the use of GenAI tools in protective orders more broadly. How do the parties intend to protect against inadvertent disclosure of confidential information in both public and non-public tools? How will confidential information be deleted at the close of the dispute? These are just some of the questions that could be answered and addressed in the parties’ protective order.

Consider prohibiting parties from uploading or disclosing confidential information into publicly available or “open” GenAI platforms

It is likely no coincidence that all of the cases we have examined to date have related to publicly available GenAI tools, where the risk of disclosure of sensitive or privileged information beyond the parties or their counsel is heightened. Companies should consider incorporating explicit GenAI-related provisions that prohibit uploading confidential information to publicly available GenAI tools that retain data or train models.

Consider security, retention and privacy considerations around the use of GenAI tools

We have previously discussed the importance of cybersecurity, data breach notice provisions, privacy and deletion or retention-related provisions in protective orders, particularly where discovery of non-US custodians or sensitive personal information may be implicated. These same considerations apply in the GenAI context. Parties should consider whether to include provisions that allow them to obtain assurances that confidential information disclosed to GenAI tools is afforded similar protection.Companies should also consider what contractual assurances they have in place with any permitted GenAI providers who process confidential information addressing these issues.

Consider existing provisions which may apply to GenAI use

Regardless of whether GenAI is specifically mentioned within an existing protective order, the parties likely still have existing obligations regarding the use, disclosure and protection of confidential information that may also apply in the GenAI context. Companies may want to consider how their existing protective orders may be interpreted in light of recent GenAI rulings, and how GenAI is currently being used to review or produce information governed by the parties' protective order.

Incorporating GenAI-specific protective order provisions may minimize the unique risks associated with GenAI tools.

We will continue to monitor developments and update our analysis as courts, bar associations and rule makers provide additional guidance. Our team has extensive experience dealing with GenAI from both a governance and litigation perspective. For any questions regarding issues relating to GenAI governance or litigation matters, please contact us.

Footnotes

  1. 1

    United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026).

  2. 2

    Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026).

  3. 3

    * *Archie Morgan v. V2X, Inc.**, No. 25–CV–01991–SKC–MDB, 2026 WL 864223 (D. Colo. Mar. 30, 2026).

  4. 4

    Warner, 820 F. Supp. 3d 629, at 634; see also Morgan, 2026 WL 864223, at *3-5.

  5. 5

    Id. at *636 (emphasis in original).

  6. 6

    Morgan, 2026 WL 864223, at *5.

  7. 7

    Id. at *7.\

  8. 8

    In re OpenAI, Inc., Copyright Infringement Litig., No. 25-MD-3143 (SHS) (OTW), 2026 WL 21676, at *2 (S.D.N.Y. Jan. 5, 2026).\

Contact