What Clients and Attorneys Need To Know
From: Whitney, Thompson, & Jeffcoach (WTJ)
What you type into ChatGPT does not stay there. Like emails, text messages, and other documents, AI prompts, outputs, and chat histories can become discoverable and ultimately end up in a courtroom. As attorneys and clients increasingly rely on generative AI to draft documents, summarize records, and analyze legal issues, many assume those conversations remain private. Recent court decisions suggest otherwise.
For clients, United States v. Heppner (United States v. Heppner, 820 F.Supp.3d 292 (2026)) addressed whether communications with a generative AI platform are protected by the attorney-client privilege. There, the defendant, in a criminal proceeding, uploaded confidential legal information and case-related documents to Anthropic’s Claude to assist with his defense. The government sought access to those AI-generated documents, and the court held they were not protected by the attorney-client privilege, explaining that “disclosing information to a third party generally destroys the confidentiality necessary for attorney-client privilege.” Since Claude was not acting as an attorney, the disclosure destroyed the confidentiality necessary to preserve the privilege. The court also rejected the argument that later providing the AI-generated materials to defense counsel restored the privilege.
In re OpenAI, Inc. (In Re OpenAI, Inc., 802 F.Supp.3d 688 (2025)), likewise demonstrates that AI-related documents may become the subject of discovery. There, plaintiffs sought production of a spreadsheet containing references to AI outputs and prompts, log entries, Slack messages, and emails that OpenAI had withheld as privileged. The Southern District of New York rejected many of those privilege claims and ordered production of the portions “not protected by attorney-client privilege,” explaining that communications do not become privileged merely because they were created at the direction of in-house counsel. The decision demonstrates that AI-related documents may be subject to production when they are relevant and not protected by privilege.
Concord Music Group, Inc. v. Anthropic PBC (Concord Music Grp., Inc. v. Anthropic PBC, 772 F.Supp.3d 1131 (2025)), however, illustrates that AI work-product is not automatically discoverable. There, Anthropic sought production of AI prompts, outputs, and account settings generated during plaintiffs’ pre-suit investigation, including materials not relied upon in the complaint. The Northern District of California agreed that attorney-crafted prompts constitute attorney work product because they reveal counsel’s mental impressions and litigation strategy. But because plaintiffs had already produced the prompts and outputs on which they relied, the court held any waiver was limited. Applying the Ninth Circuit’s fairness principle, the court explained that “the scope of [a work product] waiver must be ‘closely tailored … to the needs of the opposing party,'” and denied Anthropic’s broader request for all undisclosed prompts and outputs.
California’s discovery statutes are broad enough to reach AI-related materials when relevant and not privileged. Code of Civil Procedure section 2017.010 permits discovery of any nonprivileged matter relevant to the action, and sections 2031.010 et seq. authorize the inspection and production of electronically stored information. This includes AI prompts, outputs, chat histories, and account settings, provided the request reasonably identifies the information sought. (Code Civ. Proc., § 2031.030, subd. (a)(1).)
Although California courts have not yet addressed these issues, the trend is clear: AI conversations are increasingly treated like other forms of electronically stored information. Attorneys should assume AI-generated materials may be discoverable and may reveal information directly relevant to the claims and defenses at issue. When drafting requests for production, attorneys should consider defining “documents” to include AI prompts, outputs, chat histories, questions, answers, and other AI-generated communications. Consistent with Concord Music Group, Inc., such requests should be narrowly tailored to the needs of the case, while recognizing that In re OpenAI, Inc. and Heppner demonstrate that AI-related documents may be subject to production when they are not protected by privilege. Likewise, clients should avoid uploading confidential legal advice, pleadings, or sensitive case facts into consumer AI platforms without consulting counsel. In addition, everyone should think twice before inputting any and every thought that comes to mind, and having AI analyze every whim of their mind’s inner workings. Even if a question is posed innocently, that question and conversation thread can later be spun by savvy litigators to fit into their preferred narrative of a case. Thus, what may seem like a simple AI conversation today could later become Exhibit “A.”
