Your client receives a complaint or demand letter. Before they call you, they upload it to ChatGPT and ask: What are my best defenses? From there, the client continues to use generative AI on their own to evaluate discovery, test arguments, and draft communications with the counterparty, all without telling you.
Naturally, opposing counsel wants all your client’s generative AI inputs and outputs in discovery. Must they be produced?
The basic Rule 26(b)(3) work product framework protects (1) documents and tangible things, (2) prepared by or for a party or the party’s representative, (3) in anticipation of litigation or for trial.
On its face, that familiar framework seems adaptable to AI-assisted litigation preparation. But the first cases to address the issue reveal some nuanced questions: whose AI-assisted litigation preparation qualifies for protection, and who counts as a party’s representative?
Courts have recognized work product protection where pro se litigants used generative AI to help prepare their own employment cases. A more recent Massachusetts decision, however, declined to extend that protection where a represented plaintiff’s nonlawyer girlfriend performed the AI-assisted work in a business dispute.
Between these cases remains an unresolved and increasingly important question for business litigators: whether a represented party’s own, independent use of generative AI to prepare for litigation is protected as work product. The developing case law has important implications for how business litigators advise their clients.
The first principle to emerge from the cases is that, generally, a self-represented litigant’s own AI-assisted litigation-preparation work can be protected as work product.
In Warner v. Gilbarco, Inc., an employment case from the Eastern District of Michigan, the pro se plaintiff used ChatGPT to assist him with his litigation, and the defendant moved to compel production of the AI-assisted materials.
The court held that the plaintiff could assert work product protection over those materials, reasoning that ChatGPT and similar programs are “tools, not persons,” and that the defendant was essentially seeking the pro se litigant’s analytical and drafting process in anticipation of litigation, materials traditionally protected by the work product doctrine.
The first cases to address the issue reveal some nuanced questions: whose AI-assisted litigation preparation qualifies for protection, and who counts as a party’s representative?
Similarly, in Morgan v. V2X, an employment case from the District of Colorado, the court acknowledged that work product protection can extend to a pro se litigant’s AI use, reasoning that a pro se litigant plays the part of both a party and an advocate, and that the power of AI “make[s] that dual role surmountable” for unrepresented litigants. The court did not grant blanket protection, however: it ordered the plaintiff to disclose which AI tool had been used and imposed guardrails on the plaintiff’s AI use going forward.
A second principle to emerge from the recent case law is that use of a third-party AI platform does not necessarily waive work product protection. Courts have held that waiver occurs only where the disclosure is made to an adversary, not merely to any third party.
The Warner and Morgan courts reasoned that AI platforms are tools, not adversaries, and that the materials were unlikely to reach the adversary absent some mechanism for obtaining them, such as legal process.
Shealy v. Seaside Investments, a recent decision from the Massachusetts Superior Court‘s
Business Litigation Session, comes out the other way. There, the girlfriend of a represented plaintiff used ChatGPT to prepare a response to a demand letter and other materials related to the dispute. When the defendant sought to compel disclosure of those materials, the court rejected the plaintiff’s work product arguments and allowed the defendant’s motion.
The court reasoned that neither the ChatGPT prompts nor the AI-generated output had been prepared by a qualifying “representative,” explaining that the girlfriend’s queries and the resulting output would be shielded from disclosure only if performed at the request of plaintiff’s counsel.
In reaching that conclusion, the court distinguished Warner and Morgan, both of which involved pro se litigants who personally used generative AI in preparing their own cases — circumstances in which there is no “gap between the party and the attorney” because a “pro se litigant is simultaneously the party and the advocate.”
But the distinction between Warner and Morgan, on one hand, and Shealy, on the other, turns not on whether the litigant is represented by counsel, but on whether the person who used generative AI is a qualifying “representative” of the party, or someone acting under a qualifying representative’s direction, under Rule 26(b)(3), which protects qualifying materials prepared “by or for another party or its representative.”
In Shealy, the plaintiff himself did not interact with ChatGPT; his girlfriend did. Thus, Shealy does not necessarily establish that representation itself defeats work product protection for independent AI-assisted litigation preparation. The materials at issue were not prepared by the party, but by a third person whom the court did not treat as a qualifying “representative” within the meaning of Rule 26(b)(3).
Whether a represented party’s independent interactions with generative AI are protected by Rule 26(b)(3) is a question that remains unanswered in the civil context. In that circumstance, the materials would at least facially satisfy the rule’s requirement that they be prepared “by … another party … .” On the other hand, the absence of direction by counsel arguably creates a sufficient “gap between the party and the attorney” to preclude application of the work product doctrine.
Rather than guess how a court might answer this question, the prudent lawyer will avoid the question altogether. Address your client’s use of generative AI at the outset of your engagement. Tell your clients that independent use of AI in connection with the litigation may create material that has to be produced in discovery. Put a provision in your engagement letter that directs the client not to use AI regarding the case unless counsel specifically instructs them to. Include the same provision in any subsequent litigation hold notices.
Generative AI is novel. But the first wave of civil cases suggests that courts are largely addressing it through longstanding principles. The difficult questions concern who created the AI-assisted material, at whose direction, and whether that person and that process fall within the categories protected by the work product doctrine.
For business litigators, the immediate concern is what clients — and everyone assisting them — are doing with generative AI outside counsel’s supervision. Addressing client AI use early can help avoid a discovery headache later.
Matthew A. Kane is Smith Kane’s managing partner and a member of the Boston firm’s management team. He represents banks, financial institutions, businesses and individuals as a member of the firm’s business litigation practice.