An Overview of GenAI Issues in the Courtroom

by Taylor Galdi | May 27, 2026

Photo by Emiliano Vittoriosi via upsplash

An Overview of GenAI Issues in the Courtroom

Since the proliferation of generative artificial intelligence (GenAI) tools such as ChatGPT, judges have been forced to consider the appropriate role of these tools in the courtroom. More often than not, questions regarding GenAI in the courtroom arise not due to judges’ personal interest in the topic, but because lawyers and litigants alike are supplying the court with motions, briefs, and other filings that are later determined to have been prepared using the assistance of GenAI.

Some of the potential concerns surrounding the use of GenAI to prepare court documents may be fairly obvious, especially given some of the more parabolic, audience-conscious judicial opinions that have received coverage in the mainstream media.^[1] In the typical case, a party’s use of AI is only detected when opposing counsel or the court attempts to look up a citation contained within a document, only to find out that the citation does not exist (or, at the very least, is not legally relevant) because GenAI “hallucinated” the cited case.^[2] In response to these mistakes, courts may decide to admonish or sanction the party,^[3] or, more broadly, issue standing orders requiring that parties disclose their use of GenAI in that particular judge’s courtroom.^[4] However, in the absence of more uniform, clearly-articulated standards, individual judges are mainly relying on a reactive, ad hoc approach to instances of GenAI usage upon detection.^[5]

Both criminal and civil pro se litigants—who are responsible for 50% of the cases filed in the U.S. Court of Appeals^[6]—may be particularly inclined to turn to GenAI as they attempt to navigate the confusing world of litigation without representation. In the civil context, many judges respond to the use of GenAI in motions and filings by invoking Federal Rule of Civil Procedure 11 (or its state law analogue) to remind lawyers and pro se litigants of their obligations to ensure that the legal contentions contained within a pleading, motion, or other paper are “warranted by existing law,” and to remind parties of the consequences if they fail to do so.^[7] Indeed, in a sanctions order against a law firm that used ChatGPT in its filings, a federal district court judge in New York lamented that “[a] fake opinion is not ‘existing law’... [a]n attempt to persuade a court or oppose an adversary by relying on fake opinions is an abuse of the adversary system.”^[8]

However, by its terms, Rule 11 does not cover GenAI usage in the context of discovery.^[9] A recent case decided in the District Court for the Eastern District of Michigan, Warner v. Gilbarco, Inc., presented the first occasion for a court to determine the discoverability of a pro se litigant’s communications with ChatGPT.^[10] It is worth thinking through the issues this case raises, as it implicates questions regarding the policy considerations governing the discovery rules in civil cases and, specifically, how policies around pro se litigants’ use of GenAI might be used to create more fairness in an imbalanced adversarial system.

The Issue in Warner v. Gilbarco, Inc.

Although Warner v. Gilbarco, Inc. is a novel case for the aforementioned reasons, the underlying dispute was a fairly routine claim of employment discrimination and retaliation. In 2024, Sohyon Warner sued her former employer, Gilbarco, Inc.. Though Warner was initially represented by counsel, in a subsequent order, the court alluded to “a material breakdown in the attorney-client relationship in terms of communication and trust,” which caused Warner to file a Motion to Withdraw Attorney.^[11] As such, Warner represented herself as a pro se litigant for a substantive portion of her case. During discovery—at which point Warner was already representing herself—Defendants filed a motion to compel production, seeking “all documents and information concerning [Warner’s] use of third-party AI tools in connection with [the] lawsuit.”^[12] Warner objected, asserting work-product privilege. Specifically, Warner argued that she used ChatGPT as a “drafting tool ... solely to assist with drafting text,” and that Defendants “cannot discover what I think; that’s my thought process.”^[13]

The court denied Defendants’ motion, holding that Warner’s conversations with ChatGPT were privileged under Federal Rule of Civil Procedure 26(b)(3)(B)’s protection of “opinion” work product, which Warner, as a pro se litigant, was permitted to assert.^[14] The court’s reasoning is noteworthy for two reasons: (1) more generally, preventing the discovery of a pro se litigant’s opinion work product expands Rule 26(b)(3)(B)’s protections beyond the plain language of the Rule’s text, aligning its protections with the intent of the Rule; and (2) specifically, protecting a pro se litigant’s conversations with ChatGPT under the opinion work-product doctrine extends the scope of this protection for pro se litigants in a way that, to date, no other court has done.

The Importance of Protecting Opinion Work Product in an Adversarial System

In Hickman v. Taylor, considered to be the precursor to Rule 26(b)(3)(B)’s codification of the work-product doctrine, the Supreme Court first made the distinction between discoverable, nonprivileged information and an attorney’s protected work product. The Court differentiated between nonprivileged, relevant factual information contained within a lawyers’ files, which could be discoverable if helpful to an adversary’s case,^[15] and, on the other hand, the “private memoranda and personal recollections prepared or formed by an adverse party's counsel in the course of his legal duties,”^[16] which should be shielded from disclosure.

Indeed, the reasoning in Hickman v. Taylor demonstrates the policy concerns undergirding heightened protection for “opinion” work product. The Hickman opinion begins by stating that “[m]utual knowledge of all the relevant facts gathered by both parties is essential to proper litigation.”^[17] The purpose of the adversarial legal system, and the liberalizing of the modern discovery rules therein, is to avoid cases in which a party wins merely because it has pulled a “dark veil of secrecy over all the pertinent facts.”^[18] If parties were allowed to claim privilege in all cases simply to avoid disclosing advantageous information to their opponent, then a lawsuit “becomes more of a battle of deception”^[19] than a “battle of wits.”^[20]

However, the Hickman Court clarified that, despite the fact that modern discovery rules are designed toensure that cases are decided on their merits,, this purpose would not be served by allowing a party to discover the mental impressions and conclusions of the opposing party. “[I]t is essential that a lawyer work with a certain degree of privacy, free from unnecessary intrusion by opposing parties ... [t]hat is ... the necessary way in which lawyers act within the framework of our system of jurisprudence.”^[21]

Indeed, in the famous concurring opinion to Hickman, Justice Jackson summarizes this concern succinctly by stating that “[d]iscovery was hardly intended to enable a learned profession to perform its functions either without wits or on wits borrowed from the adversary.” Accordingly, a party’s attempt to discover opposing counsel’s private mental impressions not only fails to serve the policy purposes of the liberal discovery rules, but actually “contravenes the public policy underlying the orderly prosecution and defense of legal claims.”^[22] Emphasizing the extent of this protection, the Court noted that “[n]ot even the most liberal of discovery theories can justify unwarranted inquiries into the files and the mental impressions of an attorney.”^[23]

Therefore, Rule 26(b)(3)(B)’s work-product doctrine creates an almost absolute immunity from discovery for “mental impressions” and “opinions” (so-called “opinion” work product).^[24] Even if a court requires discovery of so-called “trial preparation” material, the text of Rule 26(b)(3)(B) specifically creates a carve-out, preventing disclosure of opinion work product that may be contained within the otherwise discoverable trial preparation material.

However, the Rule’s text guarantees this heightened immunity for lawyers and party representatives, not necessarily for litigants themselves.^[25] Rule 26(b)(3)(A) specifies that Rule 26(b)(3)(B)’s scope should be read as “including the other party's attorney, consultant, surety, indemnitor, insurer, or agent.”^[26]

For example, interpreting the text of Rule 26(b)(3)(B) to not cover a pro se litigant’s notes, a district court in Alabama stated, “Rule 26 does not state that it offers this protection against disclosure to a party appearing pro se.”^[27] A district court in California similarly noted that “[t]here is no such thing as ‘pro se plaintiff work product.’”^[28] Even the Hickman Court acknowledged the possibility that protecting an attorney’s work product gives a party a tremendous advantage over an unrepresented litigant who, without such protection, “could be compelled to disclose all the intimate details of his case.”^[29]

This is true even though courts consistently rule that pro se litigants are not entitled to special treatment or exemptions from procedural standards.^[30] That is, pro se litigants are expected to follow the same discovery rules as trained lawyers, yet, under those rules, they do not receive the same work-product protections as the lawyers whom they oppose. Even apart from the court’s unprecedented protection of GenAI content, the court in Warner v. Gilbarco, Inc. more generally chooses to extend the opinion work-product doctrine beyond the minimal requirements of Rule 26(b)(3)(B)’s text.^[31]

This choice reflects a subtle, functional expansion of the opinion work-product doctrine, which allows the Rule’s protections to better reflect the practical role that pro se litigants embody in their cases. Pro se litigants functionally serve as their own lawyers. To put forth their case, a pro se litigant must “assemble information, sift what he considers to be the relevant from the irrelevant facts, prepare his legal theories and plan his strategy,”^[32] in the same way that a lawyer does when representing a client. Therefore, protecting the opinions and mental impressions of a pro se litigant furthers the Hickman policy goal of ensuring that cases are won and lost on the merits, not because of an imbalance in information resulting from the presence of an attorney.^[33] Protecting the opinion work product of pro se litigants ensures that such litigation is more properly a “battle of the wits” between parties within an adversarial legal system.

How GenAI Policies Can Expand the Scope of Pro Se Litigants’ Work Product Protections

If a pro se litigant’s opinions and mental impressions deserve protection similar to that of lawyers’, the question remains to what extent this protection should be given. While attorneys will contain their mental impressions of a client’s case within documents that are routinely prepared by attorneys, such as memorandums, briefs, and notes, this is not necessarily the case for a pro se litigant, whose opinions about their case may be expressed in more informal forms of communication. Warner’s conversations with ChatGPT are such an example and highlight the ongoing debate surrounding the scope of protection for a pro se litigant’s opinion work product.

The Warner v. Gilbarco court’s justification for its denial of Defendants’ motion to compel was that Warner’s conversations with ChatGPT were solely a reflection of her internal mental impressions and opinions of her case, not a dialogue with an interlocutor. While it may seem obvious that GenAI programs are “merely tools, not persons,”^[34] this is not an inevitable conclusion. In fact, in the recent case of United States v. Heppner, a judge in the District Court for the Southern District of New York came to the opposite conclusion, finding that a represented client’s communications with Claude AI were not protected from discovery under either attorney-client privilege or work product protections. With respect to the work product doctrine, the Heppner court reasoned that the client’s conversations with Claude AI were not protected as work product because their communications with Claude AI were not “prepared by or at the behest of counsel ... nor did they reflect [their attorney’s] strategy.”^[35] With respect to the attorney-client privilege, the court held that communications with Claude AI were not confidential, which waived any attorney-client privilege.^[36] The court found that the privilege did not apply, not only because Claude AI was a “third-party AI platform,” but also because Claude AI’s parent company, Anthropic, has a user-agreement policy that states that Anthropic “collects data on both users ‘inputs’ and Claude’s ‘outputs.’”^[37]

The Warner and Heppner courts use the same exact reasoning to reach opposite conclusions. The fact that the two courts were discussing different GenAI programs cannot alone account for this difference, as ChatGPT has similar user-agreement policies to Anthropic.^[38] The primary difference is in the two courts’ orientation towards the function of GenAI, with the Heppner court viewing GenAI as a responsive, third-party interlocutor, and the Warner court viewing GenAI as a mere “drafting environment,”^[39] which Warner used to generate her own original legal opinions. The Warner court’s view of GenAI as just a drafting environment, similar to other less controversially, and more frequently, used tools such as Microsoft Word, Grammarly, and AI-supported legal research platforms,^[40] led the court to endorse the view that compelling discovery of Warner’s conversations with ChatGPT “would nullify work-product protection in nearly every modern drafting environment.”^[41] As previously discussed, this extreme conclusion is only inevitable if one does not view GenAI platforms as “third-parties,” and if one views the difference between ChatGPT and Microsoft Word as a difference in degree, rather than in kind.

Regardless, it is hard to deny the fact that the Warner court’s protection of a pro se litigant’s conversations with ChatGPT expands the scope of work-product protections for pro se litigants beyond whichever opinions are contained in traditionally-protected mediums such as private notes or memoranda. As mentioned, it is more likely that an attorney will contain their mental impressions and opinions of their client’s case within traditionally-protected mediums. This is likely a reflection both of attorneys’ understanding of work-product doctrine and the fact that courts typically address questions of work-product privilege, and what constitutes a waiver, as it applies to an attorney’s materials.

Not only is a pro se litigant less likely to have the foresight necessary to contain their opinions of their case to traditionally-privileged mediums, it is also more difficult for pro se litigants to practice this compartmentalization. Their involvement in their case is not just as a lawyer; instead, the progression and eventual outcome of the case bleed into their personal life. Accordingly, it is not difficult to imagine that a pro se litigant would write about their case in their diary, for example.^[42] These diary entries could contain a mixture of factual recountings of the events related to the suit, the pro se litigant’s emotional response to those events, and their “opinions” and mental impressions of their legal case.

The Warner court’s observation that Defendants’ motion was a “fishing expedition” because the Defendants were not seeking any relevant, nonprivileged information contained in Warner’s conversations with ChatGPT demonstrates that the focus of a Rule 26(b)(3)(B) inquiry should not necessarily be on the medium itself. Instead, the questions should be how the medium was used by the party and how the requesting party will use the information they are seeking.

As the Hickman Court noted, an attorney’s mental impressions could be contained in their files alongside descriptions of relevant, nonprivileged facts. A party should certainly be able to compel discovery for any relevant, nonprivileged information contained within a pro se litigant’s files—including their conversations with GenAI tools such as ChatGPT. However, if those conversations with ChatGPT are solely a reflection of the pro se litigant’s litigation strategies, then their opposing party should not be able to take advantage of the existing textual loophole of Rule 26(b)(3)(B) to “borrow wits” from the pro se litigant merely because they are not represented by a lawyer.

Conclusion

Courts should use the Warner case as an opportunity to evaluate the existing state of the protection of opinion work product for pro se litigants, both in the civil and criminal contexts. Generally, courts have responded to the use of GenAI in the courtroom in a reactive, ad hoc approach when confronted with a specific instance of its usage. Such a reactive and uncoordinated approach leads to disparate policies and case outcomes, as exemplified in the different conclusions reached by the Warner and Heppner courts on the question of whether GenAI constitutes a “third party.” However, this inconsistency is avoidable. Rather than issue individual sanctions or standing orders, courts should create local rules that govern the use of GenAI tools. Such rules would apply court-wide, eliminating the inconsistency and unpredictability that predominates the current landscape surrounding GenAI in courts.

Specifically in the context of civil cases, federal courts could craft local rules in line with the outcome in Warner v. Gilbarco. A local rule that creates immunity from discovery for “opinion” work product contained within conversations with GenAI tools, but permits discovery for any factual non-privileged, relevant information contained therein, would benefit pro se litigants immensely, even if that rule were also applicable to attorneys or party representatives.

Some may argue that this type of local rule is unnecessary, as Rule 26(b)(3)(B) already protects opinion work product, even when the disclosure of other trial preparation material is compelled. But a local rule that explicitly protects conversations with GenAI tools under Rule 26(b)(3)(B) is more likely to protect pro se litigants who, for the aforementioned reasons, are less likely to contain their opinions and mental impressions to traditionally-privileged mediums.

Therefore, protecting the strategic conversations that a pro se litigant has with GenAI tools will functionally provide more protection for pro se litigants under Rule 26(b)(3)(B) than currently required by the Rule’s text. The explicit protection of opinion work product contained within conversations with GenAI would bring Rule 26(b)(3)(B) more in line with its intended policy goals, by ensuring that cases are won and lost based on a “battle of the wits.” Moreover, it would ensure that legal outcomes are not based on a party’s ability to advantageously compel discovery of an unsuspecting pro se plaintiff’s legal strategies and opinions of their case merely because the unrepresented litigant used GenAI.

^[1] Molly Bohannon, Lawyer Used ChatGPT In Court—And Cited Fake Cases. A Judge Is Considering Sanctions, Forbes (June 8, 2023).

^[2] Id.

^[3] Lacey v. State Farm Gen. Ins. Co., No. CV 24-5205 FMO (MAAx), 2025 WL 1119698 (C.D. Cal., Mar. 3, 2025).

^[4] Maura R. Grossman, Paul W. Grimm & Daniel G. Brown, Is Disclosure and Certification of the Use of Generative AI Really Necessary?, 107 Judicature 68, 77 (2023).

^[5] See Kruse v. Karlen, 692 S.W.3d 43, 52 (E.D. Mo. Ct. App. 2024) (“We regret that Appellant has given us our first opportunity to consider the impact of fictitious cases being submitted to our Court, an issue which has gained national attention in the rising availability of generative A.I.”)

^[6] United States Courts, Judicial Business 2025 (2025), https://www.uscourts.gov/data-news/reports/statistical-reports/judicial-business-united-states-courts/judicial-business-2025.

^[7] Fed. R. Civ. P. 11(b)-(C).

^[8] Mata v. Avianca, Inc., 678 F. Supp. 3d 443, 455 (S.D.N.Y. 2023).

^[9] Fed. R. Civ. P. 11(D).

^[10] Warner v. Gilbarco, Inc. (Warner I), No. 2:24-cv-12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026).

^[11] Warner v. Gilbarco, Inc. (Warner II), No. 2:24-cv-12333, 2025 WL 2985113, at *1 (E.D. Mich. Aug. 26, 2025).

^[12] Warner I, 2026 WL 373043, at *11.

^[13] Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 2026 U.S. Dist. Ct. Motions LEXIS 32805 (E.D. Mich. Jan. 6, 2026).

^[14] Warner I, 2026 WL 373043, at *11.

^[15] Hickman v. Taylor, 329 U.S. 495, 511 (1947).

^[16] Id. at 510.

^[17] Id. at 507.

^[18] Id. at 506.

^[19] Id. at 507.

^[20] Id. at 516.

^[21] Id. at 511.

^[22] Id.

^[23] Id. at 510.

^[24] Jennifer A. Gundlach & Zeus Smith, Expanding the Federal Work Product Doctrine to Unrepresented Litigants, 30 Geo. J. Poverty L. & Pol’y 49, 63 (2022).

^[25] Id.

^[26] Fed. R. Civ. P. 23(b)(3)(A).

^[27]Ross v. Sejin Am., Inc., No. 3:18-cv-537-RAH-JTA, 2021 WL 6973877, at *2 (M.D. Ala. Apr. 9, 2021).

^[28] Simmons v. Adams, No. 1:10-cv-01259-LJO-SKO PC, 2013 U.S. Dist. LEXIS 84429, at *6 (E.D. Cal. June 13, 2013).

^[29] See Hickman, 329 U.S. at 506.

^[30] Gundlach & Smith, supra note 24, at 53.

^[31] See Warner I, 2026 WL 373043 at *12.

^[32] Hickman, 329 U.S. at 510.

^[33] See Id.

^[34] Warner, 2026 LEXIS 32805.

^[35] United States v. Heppner, No. 25 Cr. 503 (JSR), 2026 WL 436479, at *3 (S.D.N.Y. Feb. 17, 2026).

^[36] Id. at 7.

^[37] Id.

^[38] Warner I, 2026 WL 373043 at *12.

^[39] Id. at *13.

^[40] Warner, 2026 LEXIS 32805.

^[41] Warner I, 2026 WL 373043 at *13.

^[42] Dowden v. Superior Ct, 86 Cal. Rptr. 2d 180 (Cal. Ct. App. 4th 1999).