Legal Discovery Zeroes In on AI Prompts as Courts Compel Production
US courts are compelling production of AI prompt data in legal discovery, making prompts central evidence.
Why it matters: Legal teams must review how AI-generated content and prompts are managed, as courts treat these records like any other electronically stored information (ESI). Mishandling AI prompt logs could expose sensitive strategy or client data in litigation.
- In January 2026, the Southern District of New York ordered OpenAI to produce 20 million ChatGPT prompt logs in a copyright case.
- Privilege claims over AI-generated materials varied in 2023-2024 cases like United States v. Heppner and Tremblay v. OpenAI.
- Courts are applying traditional e-discovery rules to AI data, often requiring wide production of prompt history.
- Counsel are advised by experts Joseph Kamelhar and Richard Weber to assume prompts, outputs, and uploads may be discoverable.
AI-powered tools are now routine in legal and business settings—and US courts are responding by treating AI prompts and outputs as ordinary electronically stored information (ESI) in discovery.
- In January 2026, the Southern District of New York directed OpenAI to turn over 20 million anonymized ChatGPT conversations, signaling that AI prompt logs can be central in litigation.
- Numerous 2023-2024 decisions show divided approaches on privilege. In United States v. Heppner, 31 AI-generated records were found non-privileged as they were shared across systems lacking adequate confidentiality, and the AI was not an attorney. However, Tremblay v. OpenAI and Warner v. Gilbarco, Inc. preserved protection for AI work product closely tied to ongoing litigation and kept confidential.
- Julie Anne Halter and Alexa Stemmler of Greenberg Traurig note courts are holding parties to e-discovery standards: "Parties must treat [AI-generated content] like any other potentially relevant ESI." (2026 analysis)
- Joseph Kamelhar and Richard Weber, writing for JD Supra, warn: "Clients should assume, for now, that prompts, outputs, uploaded documents and saved chats may later become relevant in discovery."
- In ongoing disputes, discovery requests have encompassed up to 80,000 entries from law firm AI tools, underscoring the growing scope.
- Attorney Robert Rosenberg underscores risk: "AI creates records. Records can create exposure."
The bottom line: Treat all AI interactions as potential evidence. Legal teams should proactively document procedures for handling AI-generated data, clarify privilege boundaries, and educate clients on discovery risks tied to prompts and uploads.
By the numbers:
- 20 million — ChatGPT logs OpenAI compelled to produce in SDNY, January 2026
- 80,000 — AI prompt entries requested in ongoing e-discovery at law firms
Yes, but: Privilege may still apply if AI records are prepared for litigation and kept strictly confidential, but courts use a case-by-case approach.
What's next: More judicial guidance on AI prompt privilege is expected as discovery disputes reach appellate courts in 2026.