Prompts, privilege & discovery
Assini v. Hayward
The court quashed subpoenas seeking a pro se litigant’s OpenAI prompts, uploads, outputs, drafts, and account material.
Prompts protected01
Case background
Material facts
The plaintiffs subpoenaed OpenAI for a self-represented defendant’s interactions with generative AI concerning the litigation. The defendant moved to quash and for a protective order. An amended subpoena sought (1) all prompts, inputs, uploaded materials, and corresponding outputs or generated text used to draft, revise, or generate filings, motions, sworn statements, or communications transmitted to the plaintiffs, their counsel, or the court, or filed in the action during the relevant period; and (2) all prompts, queries, uploaded documents, and corresponding outputs specifically referencing the parties and claims. (Slip op. at 2.)
02
The decision
The court’s ruling
The court applied New York’s conditional privilege for material prepared solely in anticipation of litigation. (Slip op. at 3.) It distinguished United States v. Heppner and followed Morgan v. V2X, Inc., holding that a pro se litigant’s use of AI can involve the confidential, strategy-laden iterative work product that Rule 26(b)(3) protects. (Id. (citing Morgan v. V2X at 10–11).) The court treated the requested AI material as protected litigation preparation and quashed the subpoenas.
Why it matters
What the decision means
A self-represented litigant’s materials qualify as work product when created for litigation. The decision does not protect chats created for another purpose.
Limit of the ruling. This is a trial-court ruling in a pro se setting, not a categorical rule that every conversation with an AI system is protected.
Primary reading