The subpoena asked OpenAI for “[a]ll prompts, inputs, uploaded materials, and corresponding outputs or generated text used to draft, revise, or generate filings, motions, sworn statements, or communications” in the action, a litigant’s entire ChatGPT history with the case. On June 4, 2026, a Nassau County judge quashed it. Assini v. Hayward, 2026 NY Slip Op 26086 (Sup. Ct., Nassau County June 4, 2026). The plaintiffs seeking that history were represented by counsel. The defendant whose chats they wanted, John Recchio, had no lawyer at all.
That detail is not a footnote. It is the story.
Assini is the third decision in four months to protect a litigant’s AI conversations as work product. The best write-up of the line so far is Nitesh Daryanani’s, at Midpage, which reads the cases as correcting a framing error. In its telling, the question was never whether a chatbot conversation is “privileged,” as if the bot were a lawyer, but whether sharing work product with an AI tool waives the protection. On the law, I have no quarrel with that account, and I recommend it. Where I part ways is over what it treats as incidental. Every litigant who won in these cases was pro se. The one litigant who lost, in the only decision to deny protection, had lawyers. That pattern is not a coincidence. It is the work product doctrine being forced, seventy-nine years after Hickman v. Taylor, to answer a question it has been quietly avoiding the whole time. Does the law protect the lawyer’s preparation, or the litigant’s?
Three courts, four months, one chain
The line assembled itself with unusual speed, each link citing the last.
It starts in February, in Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026). Sohyon Warner, suing her former employer without a lawyer, had used ChatGPT to help prepare her case. The defendants moved to compel discovery into her AI usage, arguing she had waived any work product protection by routing her thinking through OpenAI’s servers. Magistrate Judge Anthony Patti denied the request, and his reasons repay close reading. He took the threshold point first, quoting his own pre-AI precedent. “Plaintiff, as a pro se litigant, has a right to assert work product protection over such material.” Then came the waiver analysis. Work product waiver “has to be a waiver to an adversary or in a way likely to get in an adversary’s hand,” and “ChatGPT (and other generative AI programs) are tools, not persons, even if they may have administrators somewhere in the background.” A tool is not an adversary. The court also noted, with visible skepticism, that the defendants’ waiver theory was “supported by no case law but only a Law360 article posing rhetorical questions” and agreed with Ms. Warner that the theory “would nullify work-product protection” in ordinary modern practice.
Pause on that last point. The argument the court adopted was written by the pro se plaintiff herself, in her own response brief. We will come back to why that matters.
Seven weeks later, in Morgan v. V2X, Inc., No. 1:25-cv-01991 (D. Colo. Mar. 30, 2026), another pro se plaintiff had run confidential case information through an AI tool, and the defendant moved to find out which one and on what terms. Magistrate Judge Maritza Dominguez Braswell followed Warner expressly and added the sentence this line of cases will be remembered for. She wrote 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.” Note where that sentence begins. The opening clause, in the context of a pro se litigant’s use of AI, is not throat-clearing. It is the scope of the holding, and it is the part most easily lost when the sentence gets quoted. The privacy objection was that these tools are run by companies that collect and store everything. The court answered directly that “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.”
Then, on June 4, came Assini, where Justice Rhonda Fischer surveyed both federal decisions, found Morgan “persuasive,” applied New York’s conditional protection for litigation-preparation materials under CPLR 3101(d), and quashed the OpenAI subpoena outright.
Watch the shape of what just happened. A doctrine crossed from federal magistrates in Michigan and Colorado into a New York state commercial case in four months, citation by citation. And every load-bearing case in the chain was built on the litigation preparation of a person who could not afford or did not have counsel.
The case that lost
The chain has one broken link, and it is the most instructive case of the four. In United States v. Heppner, No. 1:25-cr-00503 (S.D.N.Y. Feb. 17, 2026), Judge Jed Rakoff held that a criminal defendant’s exchanges with Claude were not protected work product. Look at the ground. The Second Circuit’s formulation protects materials “prepared by or at the behest of counsel,” and Bradley Heppner, who had an entire defense team, produced his AI analyses, in his own counsel’s words, “on his own volition.” His lawyers conceded the documents did “affect” their strategy going forward but did not “reflect” it. Not prepared at counsel’s behest and not a window into counsel’s thinking, the documents got no protection.
Run Heppner and Morgan side by side and the inversion comes into focus. The represented defendant’s solo AI prep was unprotected because he had lawyers and didn’t act through them. The unrepresented plaintiffs’ AI prep was protected because, having no lawyers, their own preparation was the only litigation preparation there was. Under the “behest of counsel” framing, representation shrinks the protection your own work receives. Had Heppner been pro se, his Claude sessions would have looked exactly like Archie Morgan’s. They would have been strategy-laden, iterative, and prepared in anticipation of litigation by the party himself.
Morgan distinguished Heppner as a criminal case, and Assini repeated the distinction. Fair enough as far as it goes. But the deeper line the cases are drawing is not civil versus criminal. It is a question about whose preparation the doctrine exists to protect. And on that question, the pro se cases have the better of the history.
The doctrine never required a lawyer
Everyone remembers Hickman v. Taylor, 329 U.S. 495 (1947), as the case about “the work product of the lawyer.” The phrase, which the Court borrowed from the Third Circuit, does say lawyer. But the sentence the doctrine actually stands on reads, “Proper preparation of a client’s case demands that he assemble information, sift what he considers to be the relevant from the irrelevant facts, prepare his legal theories and plan his strategy without undue and needless interference.” Id. at 511. The rationale is not about the professional. It is about the adversary system. A side must be able to prepare in private, or preparation itself becomes discoverable ammunition, and “[i]nefficiency, unfairness and sharp practices” follow. In 1947, the only people generating preparation worth fighting over happened to be lawyers. The rationale and the profession traveled together, so nobody had to ask whether they were separable.
The 1970 codification quietly answered the question anyway. Rule 26(b)(3) protects materials prepared in anticipation of litigation “by or for another party or its representative.” The Advisory Committee said the quiet part in its note. The rule requires a special showing “not merely as to materials prepared by an attorney,” but as to materials prepared “by or for a party or any representative acting on his behalf.” A party. No lawyer required. The text has read that way for fifty-six years.
For nearly all of those years the extension barely mattered, because unrepresented parties rarely produced anything an adversary would want. They might have a few handwritten notes, but not memoranda of strategy, not iterative drafts, not theories of the case tested and refined across hundreds of exchanges. The protection ran to the party in theory and to the lawyer in practice, because lawyers were where the paper came from.
AI changed the production function. A pro se litigant with a chatbot now generates precisely the artifact Hickman described (information assembled, relevant sifted from irrelevant, theories prepared, strategy planned) in a form that is written down, stored on a third party’s server, and subpoenaable. Morgan’s “strategy-laden iterative work product” is not a new category. It is Hickman’s category, finally being produced at scale by people without lawyers. The doctrine didn’t change in 2026. The demographics of preparation did.
The chair counsel would have occupied
This is why the pro se pattern in these cases is the story and not a quirk of who got sued. The chatbot sits in the chair counsel would have occupied. It sits there not as a lawyer, for nothing in these decisions treats it as one, but as the surface on which an unrepresented person’s case preparation becomes visible. Subpoena a represented party’s preparation and you collide with counsel’s files. Subpoena a pro se litigant’s chatbot history and you get the same thing those files would have contained. You get the theories, the drafts, the worries, the strategy. The only difference is that there was never a lawyer for the protection to attach to. Three courts have now said the protection attaches anyway. It attaches to the preparation, because the preparation is the point.
There is something quietly moving in the details, too. The argument that AI waiver theories “would nullify work-product protection” in modern practice (adopted by a federal court, quoted up the chain) was authored by a pro se litigant in her own brief. The zone of privacy that Hickman called essential to our system of justice was extended to unrepresented people by unrepresented people, litigating against firms, without counsel, using the very tools at issue.
The fine print
Now the lawyer’s paragraph, because none of this is absolute. These are trial-level decisions (two magistrate orders and an uncorrected slip opinion), persuasive, not binding, and the slip op may yet be revised. The protection is conditional, not categorical. Morgan still ordered the plaintiff to disclose which AI platforms he had used with information designated confidential, and rewrote the protective order to restrict what could be fed into AI tools going forward. The waiver standard retains teeth. Disclosure in circumstances that substantially increase the likelihood an adversary obtains the materials still waives, and a tool whose terms volunteer your inputs to others is a different analysis than one that doesn’t. And represented parties should read Heppner as a standing warning. In the Second Circuit’s formulation, your solo AI sessions may enjoy less protection than your unrepresented opponent’s.
Nor did the courts hand pro se litigants a free pass. In the same decision that quashed the subpoena, Justice Fischer warned that the defendant’s “use of AI frustrates the litigation” — quoting Augustin v. Formula 3 Brook. Inc., 2025 NY Slip Op 51113(U) (Sup. Ct., Kings County 2025) — “and cannot go unfettered,” and referred him to 22 NYCRR Part 161, New York’s new AI rule, effective June 1, 2026, with its verification duty and sanctions backstop. Protection for the preparation; accountability for the filings. Readers of last week’s piece on this blog will recognize the shape. The same season that extended work product protection to AI-assisted preparation also delivered New York’s Part 161 and Florida’s amended Rule 2.515. The system is not choosing between sheltering AI use and policing it. It is doing both at once, and both halves arrived within a month of each other.
Seventy-nine years later
Hickman promised that our adversary system would leave people room to prepare their cases in private. For seventy-nine years that promise was, in practice, a promise to people who could afford the professional whose name the doctrine carried. It took a chatbot, and three judges paying attention to what preparation actually looks like now, to extend it to everyone else. “Work product of the lawyer” was always a slightly inaccurate name. The Advisory Committee knew it in 1970. Warner, Morgan, and Assini are what it looks like when the rest of the system finally catches up.
The authorities: Assini v. Hayward, 2026 NY Slip Op 26086 (Sup. Ct., Nassau County June 4, 2026); Morgan v. V2X, Inc., No. 1:25-cv-01991, ECF No. 65 (D. Colo. Mar. 30, 2026); Warner v. Gilbarco, Inc., No. 2:24-cv-12333, 2026 WL 373043 (E.D. Mich. Feb. 10, 2026); United States v. Heppner, No. 1:25-cr-00503, ECF No. 27 (S.D.N.Y. Feb. 17, 2026); Hickman v. Taylor, 329 U.S. 495, 510–11 (1947); Fed. R. Civ. P. 26(b)(3)(A) and advisory committee’s note to the 1970 amendment; CPLR 3101(d); 22 NYCRR Part 161 (eff. June 1, 2026). Quotations were verified against the slip opinion and the orders as filed on the courts’ dockets.