On June 4, 2026, in Assini v Hayward, 2026 NY Slip Op 26086, a Nassau County Supreme Court (Fischer, J.) quashed non-party subpoenas seeking a litigant’s artificial intelligence prompts, uploads, and outputs, holding that AI-assisted drafting can qualify as protected litigation-preparation material.

Procedural Posture and What Was Sought

Plaintiffs served a non-party subpoena and an amended subpoena on an AI provider for materials tied to pro se defendant John Recchio’s accounts. The subpoenas sought all prompts, inputs, uploaded materials, and corresponding outputs used to draft, revise, or generate filings, motions, sworn statements, or communications transmitted in the case. They also sought prompts, queries, uploaded documents, and outputs referencing the parties, Alpha Tech Lending LLC, the claims or defenses, or filings and correspondence generated for the action. Recchio moved to quash and for protective relief; plaintiffs opposed and cross-moved to so-order the subpoena and for sanctions and other relief.

Disposition

The court granted Recchio’s motion to quash the subpoena and amended subpoena, denied Recchio’s order to show cause for lack of jurisdiction based on noncompliance with service requirements, and denied plaintiffs’ cross-motion to so-order the amended subpoena and for other relief.

Governing New York Discovery and Privilege Framework

Citing CPLR 3101(a) and CPLR 2304, the court noted that New York’s discovery regime mandates full disclosure of all matter material and necessary, while empowering courts to supervise and, where appropriate, quash or limit subpoenas to nonparties. The court stated that the requesting party must demonstrate the discovery method will result in disclosure of relevant evidence or is reasonably calculated to lead to information bearing on the claims, and that privilege is narrowly construed, with the burden on the asserting party. The court emphasized that, given New York’s strong policy favoring full disclosure, the party asserting privilege bears the burden of proving each element of the asserted privilege or protection. Materials prepared in anticipation of litigation are conditionally protected from discovery under CPLR 3101(d) and may be discovered only upon a showing of substantial need and inability to obtain their equivalent without undue hardship.

Why the Court Treated AI Prompts and Outputs as Protected

Plaintiffs argued Recchio’s AI communications were discoverable as relevant case evidence and reasonably calculated to lead to discovery bearing on the claims and defenses. Recchio responded that the communications were privileged as litigation-preparation materials and that the subpoenas should be quashed.

Addressing competing federal authorities, the court found Morgan v. V2X, Inc. persuasive in the civil, pro se context and distinguished United States v. Heppner,[1] which plaintiffs cited to argue that AI communications were not privileged. Relying on Morgan v. V2X, Inc., the court held that a pro se litigant’s use of AI to assist litigation preparation resembles strategy-laden iterative work product and that public AI data practices do not automatically defeat protection. The court concluded that Morgan’s rationale was persuasive and that Heppner, a criminal matter, was not binding and was distinguished by Morgan.

Applying New York’s conditional work product protection standards, the court found that prompts, uploads, and outputs used to prepare filings and case communications can reflect litigation strategy and mental processes akin to classic work product and litigation-preparation protected materials, and that plaintiffs had not carried the burden to overcome that protection by showing substantial need and undue hardship. The court applied CPLR 3101(d) and, crediting the persuasive reasoning in Morgan, treated the requested AI prompts, uploads, and outputs used for filings and case communications as litigation-preparation materials not discoverable absent the required showing.

Practical Takeaways for Litigators and eDiscovery Practitioners

  1.  Framing privilege assertions for AI-assisted drafting. When asserting privilege over AI-assisted materials, practitioners may wish to articulate that prompts, uploads, and outputs were created solely in anticipation of litigation or for trial, identify their role in drafting or revising filings, and, where appropriate, describe how the iterative exchanges reflect legal strategy or mental impressions, without disclosing the protected substance. The subpoena at issue expressly targeted prompts, inputs, uploads, and outputs used to draft or generate filings and communications, which the court deemed litigation-preparation materials.
  2. Preserve protection through confidentiality and purpose documentation. Parties may wish to maintain confidential handling of AI interactions used for litigation, limit sharing to counsel and necessary agents, and contemporaneously document the litigation purpose and timing to meet the “in anticipation of litigation” standard. The court cited this as a prerequisite for CPLR 3101(d) protection.
  3. Anticipate and rebut waiver arguments tied to public AI tools. The reasoning in Morgan, adopted as persuasive, indicates that use of publicly available AI does not automatically waive protection. Parties may wish to address privacy expectations, terms of use, and the functional similarity to traditional drafting tools to help avoid waiver claims.
  4.  Substantial need and undue hardship remain the requester’s path. If an opponent claims need, advocate for particularized showings and highlight alternative sources, such as final filed documents and underlying factual materials. The court reiterated that litigation-preparation materials are protected absent a showing of substantial need and inability to duplicate without undue hardship.
  5. Scope and tailoring of third-party subpoenas. Broad, account-wide demands to AI providers risk being quashed.  Consider seeking or proposing narrowed requests tethered to nonprivileged factual information or discrete issues and assess court-ordered protocols to segregate privileged from nonprivileged data. The court exercised authority under CPLR 2304 and granted the motion to quash the subpoenas in their entirety.
  6. Discovery strategy for the requesting party. When an adversary’s use of AI is suspected, consider pursuing the underlying, nonprivileged facts, metadata of filed documents, or disclosures regarding AI’s role – rather than demanding strategy-laden prompts. If specific misrepresentations appear to be AI-generated (e.g., “hallucinated” case citations), consider building a record to show substantial need and undue hardship. In Assini v. Hayward, plaintiffs argued that relevance to the claims and defenses justified the subpoena, but their broad request for prompts and corresponding outputs was quashed.

What This Signals for Future AI Discovery in New York

Assini v. Hayward indicates that New York trial courts may treat AI prompts, uploads, and outputs as litigation-preparation materials where they are created solely for litigation – particularly in civil matters. Requesting parties may wish to calibrate subpoenas to avoid wholesale intrusions into litigation strategy and focus on nonprivileged facts, while responding parties should consider documenting purpose, maintaining confidentiality, and be prepared to substantiate the privilege claim. Courts may continue to exercise discretion to limit overbroad AI-provider subpoenas and to enforce emerging AI practice rules[2], including potential sanctions for noncompliance. The court underscored its broad discovery supervision authority, granted the motion to quash, and directed adherence to 22 NYCRR Section 161 with a sanctions warning.

[1] A full description of the Morgan and Heppner decisions can be found at Work Product Protection and the Disclosure of AI Tools in Discovery: Lessons from Morgan v. V2X (Part I).

[2] An example of rules can be found at Navigating AI Disclosure Rules in New York Courts.