Are AI prompts, and their generative outputs, discoverable in litigation? A handful of recent district court cases suggest the answer depends on whether the AI prompts and outputs constitute attorney work product.
In Tremblay v. OpenAI, Inc., 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024), the court held that AI prompts written by lawyers can constitute opinion work product when used for litigation-related purposes. The court explained that AI “prompts were queries crafted by counsel and contain counsel’s mental impressions and opinions about how to interrogate [an AI tool], in an effort to vindicate Plaintiffs’ copyrights against the alleged infringements.” In so doing, the court squarely rejected defendant’s argument that AI prompts and outputs only rise to the level of fact work product as opposed to opinion work product. That distinction is important, as opinion work product is offered near-absolute protection from disclosure whereas fact work product is discoverable upon a showing of substantial need for the materials and an inability to secure a substantial equivalent without undue hardship.
Tremblay suggests courts will favor construing AI prompts and outputs as opinion work product when used as a part of legal strategy, and at least one other court in the Northern District of California has followed suit. See Concord Music Group, Inc. v. Anthropic PBC, 2025 WL 1482734 (N.D. Cal. May 23, 2025) (agreeing with Tremblay that prompts by attorneys and outputs constitute attorney work product).
In contrast to the holdings above, courts have been more willing to deem AI content generated by non-attorneys discoverable. One such example is United States v. Heppner, No. 25 CR. 503 (JSR), 2026 WL 436479 (S.D.N.Y. Feb. 17, 2026). In Heppner, Judge Rakoff ruled that AI-generated documents created by a client of their own volition, and later transmitted to their counsel, were not protected by attorney-client privilege or the work product doctrine. Because the client acted of their own accord to generate the documents, and because the documents did not reflect the legal strategy of his counsel, the court held that the client’s AI-generated documents were not work product. The court also determined that the client’s AI-generated documents were not protected by attorney-client privilege because it considered the act of communicating with the AI tool to be an act of disclosure to a third party that broke confidentiality. In so holding, the court emphasized that the AI tool used by the client had an express provision that what was submitted was not confidential. Had the client communicated with an enterprise AI tool that maintained the confidentiality of user inputs, the outcome may have been different.
AI prompts and outputs from non-legal corporate employees have also been deemed discoverable. Concord Music Group, Inc. v. Anthropic PBC, 2025 WL 2267950, at *1 (N.D. Cal. Aug. 8, 2025). It follows to reason that AI prompts and outputs written by non-legal corporate employees may not be entitled to the same protections from disclosure as AI prompts and outputs written by legal counsel, and remain vulnerable to discovery in the current legal landscape.
The treatment of AI prompts and outputs as a form of discovery is a relatively novel area of the law. But in light of these decisions, defendants that incorporate the use of AI prompts as a part of their ordinary course of business in the Ninth Circuit should be cognizant of the possibility that AI prompts and their resultant outputs by their non-legal corporate employees could be deemed discoverable. Even when AI prompts and their outputs are used for litigation-related purposes, corporate counsel should take careful steps to treat this information as confidential to avoid waiver of their work product protection.