A federal magistrate judge in Michigan has confirmed what many practising lawyers have quietly hoped for: using an AI tool like ChatGPT to prepare litigation strategy does not forfeit work-product protection. The February 2026 ruling in Warner v. Gilbarco is the first explicit judicial statement in the United States that AI assistance in litigation sits squarely within the attorney — or litigant — work-product doctrine, not outside it.
Background and Facts
Sohyon Warner, a pro se plaintiff, filed suit against Gilbarco, Inc. and co-defendants in the Eastern District of Michigan. Like a growing number of self-represented litigants — and, increasingly, represented parties — Warner used ChatGPT to assist in preparing her litigation materials. The precise nature of her AI interactions was not fully disclosed in the public record, but the defendants became aware of her ChatGPT use and moved to compel production of her AI-related work: the prompts she entered, the outputs she received, and the extent to which she had relied on AI assistance in drafting filings and shaping litigation strategy.
The defendants’ core argument was that by routing her thought process through a third-party AI system — ChatGPT, operated by OpenAI — Warner had either waived work-product protection entirely or had rendered those materials discoverable as something other than protected work product. The argument drew on a loose analogy to waiver doctrine: just as disclosing a document to a third party outside the privilege relationship can waive attorney-client privilege, the defendants contended that feeding litigation strategy into a third-party AI tool stripped the resulting material of protection.
The magistrate judge was unpersuaded. On 10 February 2026, the court denied the defendants’ motion in its entirety.
What the Court Decided and Why
The court’s ruling rested on three distinct grounds, each independently sufficient to deny the motion.
First, the court found the defendants’ requests irrelevant and disproportionate to the needs of the case. Discovery under Federal Rule of Civil Procedure 26(b)(1) is bounded by proportionality: the burden or expense of the proposed discovery must not outweigh its likely benefit. The court found that the mechanics of how a litigant prepared her filings — whether she used a typewriter, a legal research database, or a generative AI tool — bore no relevance to the substantive merits of the dispute. Gilbarco’s desire to peer behind the curtain of Warner’s litigation preparation did not satisfy the relevance threshold, let alone the proportionality calculus.
Second, and more significantly, the court held that work-product doctrine protected the AI interactions. The work-product doctrine, codified in FRCP 26(b)(3) and rooted in the Supreme Court’s decision in Hickman v. Taylor (1947), shields materials prepared in anticipation of litigation from compelled disclosure. The doctrine’s most robust protection extends to “opinion work product” — materials that reveal a party’s or attorney’s mental impressions, conclusions, opinions, or legal theories. The court reasoned that prompts entered into ChatGPT in the course of preparing litigation necessarily reflect the litigant’s thinking about case strategy, legal theories, and factual framing. Compelling production of those prompts would, in effect, compel disclosure of the litigant’s mental impressions — precisely what the doctrine exists to protect.
Third, the court rejected the waiver argument on principle. Waiver of work-product protection through disclosure to a third party requires that the disclosure be inconsistent with maintaining confidentiality from adversaries. The court drew a sharp distinction between disclosing materials to an adversary (which waives) and using a tool — even a third-party tool — in the process of preparing litigation materials (which does not). A litigant who uses a word-processing program, a legal research service, or a cloud-based document storage platform does not waive work-product protection merely because those services are operated by third-party companies. The court treated generative AI in the same category: a tool used in the preparatory process, not a person or entity to whom confidential information is “disclosed” in any legally meaningful sense.
The Principle Established
The ruling in Warner v. Gilbarco establishes a clear baseline: the mere use of an AI tool to assist in litigation preparation does not, by itself, waive work-product protection. A litigant — whether represented or pro se — who uses ChatGPT or a comparable generative AI tool to think through legal arguments, draft motions, or develop case strategy retains work-product protection over the product of that process. Courts will not, on the defendants’ logic, treat AI as a “third party” whose involvement breaks the chain of confidentiality.
Critically, the court characterised generative AI as a tool, not a person. This framing matters enormously. Work-product doctrine has always accommodated the use of tools — libraries, research services, paralegals working under attorney supervision — without requiring that every tool interaction be disclosed. Treating AI as a sophisticated tool rather than an independent third-party participant aligns with the functional reality of how AI systems are used in legal practice today.
The decision also forecloses a line of litigation harassment that had begun to emerge: adversaries seeking discovery of AI usage as a fishing expedition into opposing counsel’s or a pro se litigant’s preparation process. After Warner, such requests face a high bar at the threshold relevance and proportionality stages, before work-product doctrine even comes into play.
The India Angle
Indian advocates who use AI tools for litigation preparation will find the Warner reasoning immediately useful, even though it originates in a United States federal court. The principle that a lawyer’s preparatory use of a tool does not strip case strategy of its protected character maps onto Indian professional ethics without difficulty. Under the Bar Council of India Rules, an advocate’s duty of confidentiality to the client is broad: information relating to a client’s case, shared in the course of the retainer, must not be disclosed. The question of whether feeding that information into a generative AI platform constitutes a “disclosure” to a third party — thereby breaching confidentiality — is one that Indian advocates are now confronting in practice.
The Warner court’s framing of AI as a tool rather than a person offers Indian advocates a principled basis for the position that using AI to draft arguments or analyse case materials does not, in itself, breach client confidentiality. However, the analysis cannot stop there. India’s Digital Personal Data Protection Act, 2023 introduces a separate layer of obligation: personal data of clients that is fed into AI systems may constitute processing of personal data, triggering obligations around consent, data localisation, and purpose limitation that do not map neatly onto common-law privilege doctrine. Indian advocates must therefore maintain a dual framework — professional privilege on one axis, data protection compliance on the other — when deciding how and whether to use AI in case preparation.
Case Details
| Case Name | Sohyon Warner v. Gilbarco, Inc., et al. |
| Court | United States District Court, Eastern District of Michigan |
| Decision Date | 10 February 2026 |
| Key Issue | Whether using ChatGPT for litigation waives work-product protection |
| Holding | Motion to compel denied; AI use does not waive work-product protection; AI is a tool, not a third party |
| Doctrine | Work-product doctrine (FRCP 26(b)(3)); proportionality (FRCP 26(b)(1)) |
| Significance | First explicit ruling that AI tool use in litigation does not forfeit work-product protection; forecloses adversarial discovery of AI interactions as a matter of litigation strategy |