AI and Work Product: What the Early Court Decisions Actually Hold

This post is for lawyers and is informational only, reflecting decisions as of June 2026. These are recent trial and district court rulings; confirm current status before relying on any of them.

A fear took hold among litigators in early 2026: that a client or lawyer using ChatGPT to help prepare a case had quietly handed the other side a discovery goldmine, prompts, drafts, and all. The early decisions are more reassuring than that, but they draw a line worth understanding, and the line is not about the tool. It is about who directed the work.

Four rulings map the terrain.

The protective decisions

In Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), the court refused to compel a party’s internal analysis and mental impressions, warning that doing so would “nullify work-product protection in nearly every modern drafting environment, a result no court has endorsed.” It also told the defendants their “preoccupation with Plaintiff’s use of AI needs to abate.”

In Morgan v. V2X, Inc., No. 25-cv-01991-SKC-MDB (D. Colo. Mar. 30, 2026), the court took a practical route: it amended the protective order to bar parties from uploading confidential information into mainstream AI tools unless the tool contractually guarantees confidentiality, no model training, and deletion on request. AI use was not forbidden; careless AI use was.

In Rehab Lending LLC v. Hayward, No. 607683/2024 (N.Y. Sup. Ct. Nassau Cnty. June 4, 2026), the court quashed a subpoena to OpenAI seeking a self-represented party’s ChatGPT prompts, inputs, uploads, and outputs. It held that AI-tool interactions used to prepare for litigation can be conditional work product under CPLR 3101(d), and that the public availability of an AI tool does not automatically waive that protection. It found Morgan persuasive and declined to follow the criminal ruling the other side relied on.

The limit

Shealy v. Seaside Investments, LLC, No. 2684CV00799-BLS2 (Mass. Super. Ct. June 2026), is the case that shows where protection runs out. There, a represented party’s romantic partner ran ChatGPT queries with no involvement from counsel. The court held that the output was not protected, reasoning that “AI is a tool not a person” and that its output does not disclose “the mental impressions, conclusions, opinions, or legal theories of an attorney or other representative of a party.” It distinguished the protective cases as resting on pro se litigants and found the self-directed, counsel-free use before it more like the criminal decision.

The through-line: counsel direction

Read together, these decisions do not say AI use is safe or dangerous. They say protection turns on the ordinary work-product question, was the material prepared by or for a representative of a party, in anticipation of litigation, at counsel’s direction, and the AI tool is neither a shield nor a trap on its own. Where counsel directs the work, protection is available. Where a represented party freelances with a consumer tool and counsel had nothing to do with it, the doctrine may not reach it. And uploading confidential or protected-discovery material into a public tool creates a separate, avoidable problem regardless of privilege.

What to do

Treat AI-assisted litigation work like any other work product:

Document the basics. Timing, purpose, the tool used, and who directed the work. That record is what the protective decisions turned on.
Use tools you control. Enterprise platforms with confidentiality terms, no-training commitments, and deletion rights, not personal consumer accounts.
Keep protected material out of public tools. Privileged communications and protected discovery should never be pasted into a tool without the right contractual terms, as Morgan makes concrete.

The protection is real. So is the discoverability of what you carelessly put in. (For the confidentiality-and-privilege side of this, see Before You Prompt, and the firm’s AI Governance practice.)

If your matter involves AI-assisted work and you want it handled so protection holds, talk to MB Legal.

This post is informational and does not constitute legal advice. The decisions discussed are recent trial and district court rulings that may be appealed, distinguished, or superseded, and none is currently indexed on CourtListener; citations are provided in full for independent verification. For guidance on a specific matter, consult a licensed attorney.