Generative AI (Artificial Intelligence) tools like ChatGPT, Claude and Gemini are now embedded in everyday corporate and legal work. Clients and lawyers use them to draft documents, brainstorm arguments and analyze issues. But a growing body of federal court decisions makes clear that courts will apply traditional privilege and work-product rules to AI interactions without creating AI-specific exceptions. The results depend on who is directing the tool, how the tool handles data and whether counsel is meaningfully involved.
These decisions create a practical roadmap for in-house and litigation teams. The key lesson: privilege and work-product protections are available for AI-assisted work, but only if counsel controls the process from the start.
Public AI Tools and Attorney-Client Privilege
The leading decision on privilege is United States v. Heppner, 820 F. Supp. 3d 292 (S.D.N.Y. 2026). The court held that a defendant’s communications with the public AI platform Claude were not protected by attorney-client privilege. The reasoning was straightforward: Claude is not an attorney, and the defendant was not seeking legal advice from counsel through the tool. Furthermore, the platform’s privacy policy defeated any claim of confidentiality because the provider collected inputs and outputs, used them for model training and permitted disclosure to third parties.
The court also held that non-privileged AI communications do not become privileged merely because they are later shared with counsel. And even where privileged information was entered into the tool, sharing it with a public AI platform constituted disclosure to a third party, risking waiver. The court rejected work-product protection as well, because the materials were not prepared by or at counsel’s direction and did not reflect counsel’s legal strategy at the time of creation.
Notably, the Heppner court suggested the result might differ if counsel directed the AI use in a way analogous to a non-lawyer professional retained to help an attorney provide legal advice to a client such as circumstances where the professional’s involvement is necessary or highly useful to the lawyer’s legal advice. See, e.g., U.S. v. Kovel, 296 F.2d 918.
A related lesson comes from In re OpenAI, Inc., Copyright Infringement Litigation, 802 F. Supp. 3d 688 (S.D.N.Y. 2025). There, an internal spreadsheet evaluating employee prompts and AI outputs was not privileged even though it was allegedly created at in-house counsel’s direction. The document bore no privilege legend, contained no attorney entries or actual legal advice and was overwhelmingly business and technical in nature. The court emphasized that each communication must be independently assessed, and that copying counsel does not automatically confer privilege. However, employee communications that actually summarize or reflect legal advice from in-house counsel may still be protected.
Work-Product Protection for AI-Assisted Work
While privilege claims have struggled in the AI context, courts have been more receptive to work-product arguments — particularly where counsel directs the process.
In Warner v. Gilbarco, Inc., 820 F. Supp. 3d 629 (E.D. Mich. 2026), the court denied a motion to compel records of a pro se plaintiff’s ChatGPT use, holding the materials were protected work product. The court reasoned that the AI interactions reflected internal drafting, analysis and thought process prepared in anticipation of litigation. Work-product waiver requires disclosure to an adversary or in a way likely to reach an adversary, and “generative AI programs are tools, not persons.” The court rejected the argument that using ChatGPT automatically waives work product, reasoning that such a rule would nullify work-product protection in nearly every modern drafting environment.
The court in Morgan v. V2X, Inc., 2026 WL 864223 (D. Colo. Mar. 30, 2026), reached a consistent result: AI interactions do not automatically waive work-product protection where disclosure to an AI tool does not substantially increase the risk that materials will reach an adversary. The court did require disclosure of the name of the AI tool used, reasoning that this does not reveal mental impressions or strategy.
Earlier decisions established that counsel-crafted prompts qualify as opinion work product. In Tremblay v. OpenAI, Inc., 2024 WL 3748003 (N.D. Cal. Aug. 8, 2024), the court held that AI prompts crafted by counsel are opinion work product because they reveal counsel’s mental impressions about how to interrogate ChatGPT in pursuit of claims. And in Concord Music Group, Inc. v. Anthropic PBC, 2025 WL 3677935 (N.D. Cal. Dec. 18, 2025), the court similarly treated prompts and settings as opinion work product — but cautioned that protection may be waived when a party places an AI investigation “at issue” in litigation. The court ordered production of prompts and outputs from a post-suit investigation that plaintiffs intended witnesses to testify about but declined a broader waiver and emphasized that any waiver must be closely tailored.
Key Takeaways
- Do not input privileged communications or confidential client information into public AI tools. Courts will treat this as disclosure to a third party, and privilege will likely be waived.
- Use closed or enterprise AI systems with contractual confidentiality protections that prohibit the provider from storing, training on or disclosing inputs to third parties.
- Bring AI use under counsel’s direction and document that direction. Courts are more likely to protect AI interactions where counsel directs the use and controls the use of the tool, so confidentiality is preserved.
- Companies should consult with their legal and e-discovery counsel on appropriate AI use, including what may and may not be entered into AI tools regarding privileged communications, work product and confidential materials.
For more information or questions, contact Tim O’Shea.
Featured Professionals
Professionals
612.492.7373
