Courts have been scrambling to catch up with the fast and sometimes unpredictable evolution of lawyers’ use of generative AI. Many if not most courts require lawyers to advise them if they relied on AI in preparing filings and prohibit any so-called “hallucinations” (inaccurate or nonsensical outputs). Any court requirement that lawyers keep a record of prompts they used when relying on AI will implicate work product protection issues. Under what is frequently called the Sporck doctrine (from Sporck v. Peil, 759 F.2d 312, 315 (3d Cir. 1985)), courts often protect a lawyer’s selection of intrinsically unprotected things — witnesses important enough to interview; documents selected when preparing a client to testify. For instance, even though case law is not privileged or work product-protected, a lawyer’s compilation of helpful (or harmful) cases obviously would be protected from the adversary’s discovery.
Interestingly, in Tremblay v. OpenAI, Inc., Case No. 23-cv-03223-AMO, 2024 U.S. Dist. LEXIS 141362 (N.D. Cal. Aug. 8, 2024), one of the first judicial analyses of work product protection in the AI setting comes from the case in which copyright owners accused OpenAI’s use of their copyright creations. Defendant OpenAI sought all of plaintiff’s test results — “including prompts and outputs that did not reproduce or summarize Plaintiffs’ works or otherwise support Plaintiffs’ claims.” Id. at *5. The magistrate judge ordered plaintiffs to produce those negative results, along with the positive results plaintiffs were using in advancing their claims against OpenAI. U.S. District Judge Araceli Martintez-Olguin reversed the magistrate judge’s order, holding that plaintiffs’ “prompts were queries crafted by counsel and contain counsel’s mental impressions and opinions about how to interrogate ChatGPT.” Id. at *7. The judge held that (1) plaintiff had waived its work product protection for prompts that supported their claim against OpenAI, but (2) the waiver did not extend to work product “not disclosed in the operative complaint” (such as the prompts that did not reveal OpenAI’s improper use of plaintiff’s material). Id. at *10.
Other courts undoubtedly will be called upon to assess the fact and opinion work product protection for litigants’ prompts and inquiries when relying on generative AI.