AI Prompts And Outputs: Privilege And Work Product

Should AI prompts and outputs be awarded special privilege under the law? Discussions regarding attorney-client privilege and work product concerns for AI have become part of public discourse since Sam Altman, the CEO of OpenAI, posted on X about the possibility of an "AI privilege" under which, according to Altman, "talking to an AI should be like talking to a lawyer or a doctor."¹

But how are courts handling the concept of AI privilege? In a recent case involving OpenAI, the Southern District of New York rejected OpenAI's argument that certain summaries of prompts and outputs should be protected by the attorney-client privilege.² There, the class plaintiffs filed a motion to compel production of certain documents to which OpenAI claimed privilege protection. One of the documents was an excel spreadsheet with various entries written by OpenAI employees that describe and evaluate prompts to and output responses from a ChatGPT model. OpenAI asserted that the spreadsheet was created at the direction of OpenAI's in-house counsel in order to assist with copyright compliance measures for product outputs. Without making any comment about whether input of a prompt into a large language model destroys privilege, the court rejected OpenAI's privilege argument, noting that the document contained no actual legal advice sought or conveyed by an attorney.

Other courts have discussed the protected nature of prompts and outputs. For example, in Concord Music Group, Inc. v. Anthropic PBC, the Northern District of California concluded that prompts to and outputs from Claude that were submitted and obtained in the course of a legal investigation are protected under the attorney work product doctrine.³

There, the plaintiffs conducted a pre-suit investigation into potential infringement by Anthropic and relied upon certain prompts and outputs from their investigation in support of their allegations of infringement. The plaintiffs produced all prompts and outputs on which they relied to support their allegations. Anthropic then served broad discovery requests directed to all prompts and outputs, including those not relied upon by the plaintiffs. The Northern District of California concluded that such prompts and outputs can constitute attorney work product, citing to Tremblay v. OpenAI, Inc., No. 23-cv03223-AMO, 2024 WL 3748003, at *2-3 (N.D. Cal. Aug. 8, 2024). And the court rejected Anthropic's argument that the plaintiffs waived work product protection for unrelied-upon prompts and outputs, stating that Anthropic "calls for a sweeping subject-matter waiver for all unrelied-upon prompts and outputs and goes too far regardless of whether the prompts, settings and corresponding output are fact work product or opinion work product." Given the broad nature of Anthropic's requests requiring production of all undisclosed prompts, settings and outputs, the court denied Anthropic's motion to compel.

As issues involving AI prompts and AI outputs become more frequently litigated, privilege issues are likely to continue to arise. Companies should consider implementing AI policies and guidelines as part of a comprehensive risk management and training program and continue to use caution when involving AI in legal issues. TC's IP group regularly advises clients on these issues. Please reach out to a member of TC's IP group if you need assistance.

Footnotes

1 USPTO Memorandum, Interim Processes for PTAB Workload Management (Mar. 26, 2025).

2 In re: OpenAI, Inc., 2025 WL 2799890 (S.D.N.Y. Oct. 1, 2025) 1 2

3 Concord Music Group, Inc. v. Anthropic PBC, 2025 WL 1482734 (N.D. Cal. May 23, 2025)

The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.