Our Feed

“Generative tools may assist, but they can never replace the moral nerve that transforms thought into advocacy. Before this Court, artificial intelligence is optional. Actual intelligence is mandatory.” Mattox v. Prod. Innovations Rsch., LLC, 807 F. Supp. 3d 1341, 1354 (E.D. Okla. 2025) (Roberson, J., ordering sanctions)

Ask a litigator about AI use by their pro se opponents and you are likely to be met with heavy sighs and frustrated exclamations. Indeed, Damien Charlotin, the French data scientist and lawyer, has tracked a whopping 679 decisions by judges across the United States reprimanding or disciplining pro se litigants for passing along hallucinated citations, misrepresentations, false quotes, and outdated advice created by AI.[1] The data reveals that judges are issuing more reprimands and disciplinary actions each year since generative AI first became available to the public in late 2023.[2] Add in the fact that AI may also increase the number of filings pro se litigants can produce, and you might understand the frustration. Thomas v. Del. Tech. & Cmty. Coll., Civil Action No. 24-762-CFC, 2025 LX 576806 (D. Del. Nov. 6, 2025) (Connolly, J., granting defendant relief from the obligation to respond to pro se plaintiff’s numerous filings).

While courts have just begun to wrestle with the issue, early case law addressing AI use by unrepresented parties provides some helpful takeaways for attorneys facing a pro se party armed with AI chatbots and analysis tools.

Sanctions and discipline for AI hallucinations and misrepresentations may take time

Scrutinizing an opposing party’s citations, quotations, and arguments for inaccuracies is nothing new. However, in the age of AI, the task may involve discovering an opposing party has offered legal authority created from whole cloth. Legal authority fabricated by AI, often called “hallucinations,” implicates Rule 11 of the Federal Rules of Civil Procedure and its state analogues. These rules impose duties on attorneys and unrepresented parties to affix their signature to documents filed with the court, evidencing a “promise that every pleading, motion, and argument submitted to a court … rests on truth, law, and reason.”  Mattox, 807 F. Supp. at 1346. Should an attorney or unrepresented party violate Rule 11, courts may take a host of disciplinary actions including sanctions. Id.

However, in the context of AI hallucinations offered by pro se litigants, some courts have been wary of taking disciplinary action unless the hallucinations are particularly egregious or repetitive. See CH v. Bd., unpublished per curiam opinion of the Court of Appeals, issued June 11, 2026 (Docket No. 376696) n. 1 (declining to take disciplinary action against pro se litigant’s single hallucinated citation, but warning “excessive errors may lead to disciplinary actions in the future”); and Ruggierlo, Velardo, Burke, Reizen & Fox, P.C. v. Lancaster, No. 22-12010, 2023 LX 169983, at n. 5 (E.D. Mich. Sep. 11, 2023) (acknowledging defendant provided several hallucinated citations and issuing a warning about the availability of sanctions); but see Kruse v. Karlen, 692 S.W.3d 43 (Mo. Ct. App. 2024) (imposing sanctions against pro se litigant who submitted appellate brief with an “overwhelming majority” of citations being hallucinated by AI). Indeed, the data collected by Charlotin reveals that judges are much more likely to issue warnings than sanctions when it comes to pro se litigants. [3]

Litigators should understand, that despite the duty imposed on unrepresented parties under Rule 11 and its state analogues, there may be some leeway offered for initial mistakes or misrepresentations made by pro se litigants. In such situations, litigators should take the opportunity to build an evidentiary record for future sanctions and disciplinary actions provided the hallucinations continue.  Each filing should be scrutinized; every instance of hallucinated citations, quotations, and unsupported assertions should be logged; and the misuse of AI should be reported early and often.

Protective orders can prohibit pro se litigants from using AI to analyze confidential information

Most AI platforms have privacy policies that authorize them to collect data on the information users provide and the outputs the AI creates. These policies may also authorize the platform to share this data with third parties, including the government. Litigators with sensitive but discoverable information have expressed concerns about pro se parties using AI to analyze these documents and the lack of privacy protections. Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 2025 LX 494726 (E.D. Mich. Oct. 30, 2025) (Patti, J., ordering pro se plaintiff to not upload any confidential documents obtained in discovery into any AI platform).

Recent cases give credence to these concerns. In United States v. Heppner, 820 F. Supp. 3d 292, 299 (S.D.N.Y. 2026), the Southern District of New York found that information obtained from lawyers and subsequently uploaded into an AI platform by a client was not protected by attorney-client or work product privilege. While the Court’s decision was partially based on the client uploading the information without knowledge or direction from their lawyer, the court explained that its decision was also based on the AI’s privacy policies, which expressly state submitted information shared with it would not be kept confidential. Id. at 296.

Similarly, in In re OpenAI, Copyright Infringement Litig., No. 25-md-3143, 2026 LX 224885, at 3 (S.D.N.Y. Jan. 5, 2026), the Southern District of New York found that a user’s privacy interest in information shared with ChatGPT was not so great as to prohibit disclosure in discovery. The Court allowed discovery into 20 million chat logs and found that privacy interests would be adequately protected by subjecting the logs to an existing protective order. Id.

Litigators have reason to be concerned that, should an unrepresented party upload their client’s sensitive and confidential information to an AI platform, that information could be used by the AI platform for its own purposes or disclosed in an unrelated matter. However, litigators are not powerless to prevent this from happening.

In Warner, a pro se plaintiff heavily relied on AI to help with their case and had offered several hallucinated citations and quotations in her filings. Warner, LX 494726 at 7-9. Concerned that the plaintiff would use AI to analyze their client’s confidential personnel records, defense counsel sought and were granted a modification of the Court’s protective order, preventing the plaintiff from uploading documents marked confidential to any AI platform. Id. at 3.

The Warner case provides an example of solutions that may be available to protect confidential information. Litigators should not refrain from seeking similar protections when needed.

Relief may be available from overwhelming filings

Another often cited concern with AI use by unrepresented parties is that AI may enable the drafting and filing of numerous, unnecessary, and burdensome motions. Responding to these motions can greatly strain the time and resources of clients and litigators. Recognizing this strain, the District Court of Delaware has, in at least one instance, removed the obligation to respond to the voluminous AI enabled filings by a pro-se litigant. Thomas, LX 576806 at 3.

In Thomas, an instructor at Delaware community college sued the college claiming discrimination in violation of the Civil Rights Act of 1964. Thomas v. Del. Tech., Civil Action No. 24-762-CFC, 2025 LX 430358, at 2 (D. Del. Sep. 29, 2025). In the first three months of the litigation, the instructor filed over 49 motions with some containing hallucinated cases and quotations. Thomas, LX 576806 at 2. In response, the college moved for sanctions and moved to prohibit the instructor from filing future motions without the Court’s consent. Id. While the court declined to award sanctions during the early stages of the litigation and declined to bar the instructor from filing subsequent motions, the Court did find that the college had legitimate concerns. Id. In response to those concerns, the Court decided to relieve the college from its obligation to respond to any future filings by the instructor unless the Court ordered a response. Id.

While Thomas is yet another example of a court’s wariness of imposing sanctions for early instances of hallucinated cases, it does establish that courts understand the burdens voluminous filings can cause. Practitioners should seek relief from these burdens when appropriate.

Conclusion

AI presents both opportunities and challenges for the courts, litigants, and the legal profession. While generative AI may help unrepresented parties better access legal information and participate in the judicial process, recent cases demonstrate that AI-generated errors, misuse of confidential information, and an increase in burdensome filings can create significant issues for opposing parties and the courts alike. Fortunately, emerging case law also provides a roadmap for practitioners navigating these challenges.

When confronted with a pro se litigant who relies heavily on AI, litigators should consider several best practices:

• Keep a detailed record of AI-generated errors and misrepresentations. Early cases suggest that courts are often reluctant to impose sanctions based on isolated mistakes by pro se litigants. However, repeated hallucinated citations, fabricated quotations, and unsupported factual assertions can establish a persuasive record for sanctions or other disciplinary measures. Practitioners should therefore carefully document each instance of misuse and promptly bring serious concerns to the court’s attention.
• Use protective orders to safeguard confidential information. Because many AI platforms reserve broad rights to collect, store, and share user-submitted information, confidential discovery materials may be vulnerable if uploaded to those systems. Where sensitive information is involved, counsel should consider seeking protective-order provisions that expressly prohibit the use of AI platforms to analyze or process confidential documents.
• Seek procedural relief when AI-enabled filings become excessive. Courts increasingly recognize that AI tools can enable litigants to generate a high volume of motions and other filings with minimal effort. When such filings become repetitive, burdensome, or abusive, practitioners should not hesitate to seek reasonable procedural protections, including limitations on response obligations or other case-management measures designed to conserve party and judicial resources.

As AI becomes more prevalent in litigation, courts will continue to develop standards governing its use. Until then, the most effective response is not opposition to the technology itself, but vigilant advocacy, careful record-building, and proactive use of existing procedural tools. Attorneys who understand both the risks posed by AI and the remedies currently available will be best positioned to protect their clients while courts work to establish clearer rules for the future.

[1] AI Hallucination Cases, Damien Charlotin, https://www.damiencharlotin.com/hallucinations/?q=&sort_by=-date&states=USA&parties=Pro+Se+Litigant&period_idx=0&legal_fields=&graphs=1 (last visited June 16, 2026)

[2] Id.

[3] Of the 679 decisions identified by Charlotin, only 41 resulted in monetary sanctions against a pro se litigant that offered hallucinated citations, quotations, and unsupported propositions. While a number of cases were dismissed for issues related to AI usage, most cases resulted in formal warnings.

By Ben Kaiser

This publication is intended for educational purposes only. This communication highlights specific areas of law and is not legal advice. The reader should consult an attorney to determine how the information applies to any specific situation.