Subpoena For AI Prompts Of Pro Se Civil Litigant Denied

A pro se (self-represented) litigant has been using Artificial Intelligence (AI) to draft complaints, motions, and discovery documents. The opposing party issues a subpoena to the AI platform to obtain their adversary’s “prompts” to the AI company, arguing their AI software is not an attorney and their opponents’ “prompts” to the software are not protected by attorney-client privilege.

In such a case, are the AI “prompts” discoverable? A recent New York court said no.

In Assini v. Hayward, No. 607638/2024, 2026 WL 1677232 (Sup. Ct. Nassau Cty., June 4, 2026), a party's subpoena to Open AI, seeking an opposing party's AI "prompts," was denied. The party propounding the subpoena argued their opposing pro se litigant's AI communications with Claude (a series of large language models developed by American software company Anthropic) were not privileged or confidential, nor subject to work-product protection. However, the court found a reasonable expectation of some privacy and confidentiality when interacting with AI tools, even though a litigant may understand a third party is collecting and storing their information.

The ruling is from a lower New York State civil court, and contrasts with United States v. Heppner, 820 F.Supp.3d 292 (S.D.N.Y. 2026), where AI use by a criminal defendant in federal court (who had counsel, and used AI separate from his consultation with his counsel) was found not to be privileged. In Heppner, the Court found that AI communications with Claude were neither privileged nor confidential and not subject to work-product protection because the materials were created by the defendant “on his own volition,” not at counsel's direction and not reflecting counsel's mental impressions or litigation strategy.

New York courts will undoubtedly continue to wrestle with discoverability of litigants’ use of AI as filings by pro se litigants utilizing this technology continue to rise. We expect further decisions on this issue by the New York’s Appellate Divisions, as well as further decisions in federal court, both in the criminal and civil context.

In Assini, the plaintiff propounded a subpoena directed to non-party OpenAI OpCO, LLC seeking: “1. All prompts, inputs, uploaded materials, and corresponding outputs or generated text used to draft, revise, or generate filings, motions, sworn statements, or communications that were transmitted to the plaintiffs, plaintiffs' counsel, the court, or filed in this action during the relevant time period; and 2. All prompts, queries, uploaded documents, and corresponding outputs specifically referencing: (a) plaintiffs Fred Assini or Jason Napolitano; (b) Alpha Tech Lending LLC; (c) the claims or defenses asserted in this action; or (d) filings, motions, or correspondence generated for use in this action” as associated with the accounts registered to defendant Recchio.

The plaintiff asserted that the materials squarely related to evidence and the pro se defendant Recchio's factual assertions regarding the claims and defenses at issue in this litigation. Recchio's AI interactions, the plaintiff argued, were “relevant evidence or [are] reasonably calculated to lead to the discovery of information bearing on the claims,” and are therefore discoverable, citing Mendives v. Curcio, 174 A.D.3d 796, 797, 105 N.Y.S.3d 513 (2d Dept. 2019).

In his moving papers, pro se defendant Recchio contended that the subpoena and amended subpoena should be quashed, and a protective order issued, as the materials sought are privileged as “litigation-preparation.”

The plaintiff cited to United States v. Heppner, 820 F.Supp.3d 292 (S.D.N.Y. 2026), for the proposition that defendant Recchio's use of AI is not privileged. In Heppner, the Court found that AI communications with Claude were neither privileged nor confidential and not subject to work-product protection because the materials were created by the defendant “on his own volition,” not at counsel's direction and not reflecting counsel's mental impressions or litigation strategy.

However, in contrast to the criminal case United States v. Heppner, a federal magistrate judge in the Eastern District of Michigan reached the opposite conclusion in the civil case Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), denying a motion to compel discovery of a pro se plaintiff's use of AI tools such as ChatGPT and holding that such materials are protected under the work-product doctrine despite the third-party operator potentially having access.

The court in Assini referenced Warner by citing an even more recent a civil case, Morgan v. V2X, Inc., 2026 WL 864223, *5, 2026 U.S. Dist. LEXIS 67939, *10-11 (Dist Colorado March 30, 2026), where the court found that “in the context of a pro se litigant's use of AI to assist with their litigation preparation, the use of AI closely resembles the kind of confidential, strategy-laden iterative work product that Rule 26(b)(3) was designed to protect. In other words, given 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.”

The Assini court concluded the Morgan court directly addressed Heppner and found that it was not binding, as Heppner involved a criminal matter and not a civil matter.

The Assini court, quoting the Morgan court, further found that, “[h]ere, like in Warner v. Gilbarco, Inc., 820 F.Supp.3d 629 (E.D. Mich. 2026), the plaintiff can assert work product protections in connection with his AI use. It is true that AI systems like ChatGPT, Claude, Gemini, and others widely available to the public, collect user data for training and other purposes. But in this Court's estimation, that does not eliminate all expectations of privacy or automatically waive protections.” 2026 WL 864223, *5, 2026 U.S. Dist. LEXIS 67939, *10-11 [Dist. Colorado March 30, 2026].

The Assini court found the rationale in Morgan persuasive and accordingly, the subpoena and amended subpoena were quashed. Notwithstanding same, the pro se defendant’s use of AI did not go unnoticed by the court.

As of June 1, 2026, New York courts have adopted 22 NYCRR §161, “Use of Artificial Intelligence Technology,” which establishes rules for the use of AI tools by attorneys and parties (including pro se litigants) in preparing court papers in all New York State Unified Court System courts. Attorneys and parties may use AI tools in preparing court papers, provided they comply with existing duties and responsibilities for submissions. Users of AI must understand the tool’s capabilities and limitations, including the risk of generating fabricated information or fictitious citations (hallucinations). Even pro se litigants have been directed to 22 NYCRR §161 and mandated to affirm that, when submitting legal papers to the court, they are certifying that the papers contain no false materials, facts, or frivolous legal arguments. Repeated failure to abide by this provision may provide the basis for sanctions.

As such, the pro se defendant Recchio was referred to 22 NYCRR § 161 and directed to be governed accordingly. The Assini court noted that, while 22 NYCRR §161 does not prohibit the use of AI, the pro se defendant's “use of AI frustrates the litigation, and cannot go unfettered. The Assini court cited Augustin v. Formula 3 Brook. Inc., 86 Misc.3d 1236(A), 236 N.Y.S.3d 586 (Sup. Ct. Kings County 2025), another recent lower state court decision in which a litigant was sanctioned for improper use of AI.

Though the subpoena for his AI prompts was denied, the Assini court admonished the pro se litigant that continued failure to abide by New York Court Rules regarding the use of AI may result in sanctions.

Key Takeaways

For client, carriers, and litigators alike, the keytake aways are focus and persistence. Ultimately, this means more demands and more motion practice, but narrowly tailored to avoid anticipated challenges. Litigators filing motions and propounding such requests and subpoenas must note: 

• AI is a tool, not a lawyer. Whenever one shares information with a third-party AI platform, they are arguably failing to maintain the information’s confidentiality. Said materials are not necessarily created at the direction of counsel and do not necessarily qualify as protected work product. Clients and carriers should be guided accordingly.
• Litigators must assert privilege for their own AI-assisted drafting. When using AI-assisted services/materials, litigators must assert that the prompts, uploads, and outputs were created solely in anticipation of litigation or trial and/or reflect legal strategy or mental impressions, to avoid disclosing the protected substance.
• Demands for materials should be narrowly tailored and focused.

  a) Narrow the scope of AI-related discovery demands. Narrow down document productions, for analysis, data, or use of AI to formulate opinions, the AI prompts and logs used in this methodology are discoverable. 

  b) Demand an adverse expert’s use of AI litigation tools. When an expert witness uses artificial intelligence as part of their analytical workflow, the AI prompts, queries and related materials may be discoverable. In Conservation Law Foundation, Inc. v. Shell Oil Company, 628 F. Supp. 3d 416 (D. Conn. 2022), the court compelled production of AI prompts and queries used by a testifying expert, treating them as part of the expert's discoverable methodology under Federal Rule of Civil Procedure 26.

  c) Issue Demands for Specific AI tools. If specific AI tools are suspected of being used (e.g., ChatGPT, Claude, internal corporate AI) issue demands specific to those platforms and prompts.

  d) Demand an Affirmation Per 22 NYCRR §161. If false AI case citations, commonly known as “hallucinations,” are suspected, demand an affirmation from the opposing party certifying that their papers do not contain false materials, facts, or frivolous legal arguments.

• Issue a Notice to Preserve. Demand the preservation of AI prompts, logs, and associated metadata from opposing parties. In federal court, do this during the initial Rule 26(f) discovery conference. In New York State courts, a Notice to Preserve can accompany a defendant’s Answer and other initial Combined Discovery Demands.
• Focus third-party subpoenas to nonprivileged information. Broad, account-wide demands in any case risk being quashed. Make narrowed requests for: a) to nonprivileged, factual information; or b) specific issues; and c) follow court-ordered protocols to segregate privileged from nonprivileged information.
• Focus on when AI was used. Pre-existing documents that are not privileged do not become privileged because a party hands them to an attorney. Sharing unprivileged documents with AI services or generating unprivileged AI-Generated content does not retroactively shield them. Tailor discovery demands accordingly.
• Focus on who generated the materials. Demand the identity of all persons generating the materials. Using persons outside the ligation team means these documents are arguably not privileged.
• What to do when “hallucinations” are spotted. If “hallucinations” or misrepresented citations in an opposing party’s litigation documents are found, bring it to the attention of the court and file the appropriate papers. Be case specific: outline the time, costs, lack of candor, need for additional investigation of the materials, etc. Go on the attack for more information. At the court’s discretion, costs and sanctions may be awarded.

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.