Case tracker · updated July 2026
The AI Privilege Docket.
Courts are deciding, case by case, when AI conversations are protected and when they are discoverable. We track the decisions here in plain language, because your clients are not going to stop using AI while the law catches up.
Warner v. Gilbarco, Inc.
A self-represented plaintiff used a consumer AI tool to help prepare her own filings. The defense moved to compel everything about that use. The court refused, holding that generative AI programs are tools, not persons, so running material through one is not disclosure to an adversary and does not waive work product. A person representing themselves is both party and advocate, and their preparation stays protected.
Work product upheldUnited States v. Heppner
One week later, the opposite result on different facts. A represented defendant used the consumer version of a public AI tool, on his own initiative, to generate roughly thirty documents mapping his defense, then shared them with his lawyers. Counsel had not directed the work. The court ordered the material produced. Neither privilege nor work product applied: the tool is not a lawyer, its consumer terms let the provider retain and disclose what was typed into it, and the work was not done at counsel's direction. The court noted that an enterprise deployment with real confidentiality terms would present a materially different question.
No privilegeMorgan v. V2X, Inc.
A self-represented litigant's AI-assisted preparation was protected as work product. But the court entered a protective order restricting further use of consumer tools on discovery material, and required the party to disclose which tool had been used. Protection for the content, guardrails on the channel.
Content protectedConservation Law Foundation v. Shell Oil
A different context, a different answer. Prompts a testifying expert gave to an AI tool were held discoverable under the rules that govern expert disclosures, which reach further than the work-product protection that shields a party's or a lawyer's own preparation. Who is holding the tool decides the outcome here too.
Prompts discoverableAssini v. Hayward
The opposing party did not ask the litigant for his AI history. It subpoenaed the AI provider directly for his prompts and the outputs. Because the litigant was self-represented, the court held the material was work product and quashed the demand. The lesson is the subpoena itself: an AI provider holds a retained, attributable record of a party's thinking that a web search never left behind.
Subpoena quashedABA Formal Opinion 512
Not a court decision, but the backdrop to all of them: lawyers using generative AI must maintain competence with the tools, protect client confidentiality, and supervise AI-assisted work product before relying on it. Supervision is not optional, and it cannot be delegated to the software.
The split. Heppner and Warner came down the same month and cut opposite ways, and no appellate court has yet said which framework governs. The first ruling on attorney-directed AI use inside a representation, the question every case above circles and none has squarely answered. State bar guidance following Opinion 512. And discovery fights over AI vendors' retention logs, where the provider's own data practices become the evidence.
Plain-language summaries of public decisions and guidance, for information only. This is not legal advice. Corrections are welcome: we would rather be accurate than first.
The line the courts are drawing is the room we built.
CounselRoom is in an invite-only pilot for small and midsize firms.