Court Protects AI Prompt Testing as Work Product in Copyright Suit | Tremblay v. OpenAI, Inc.
Summary
Tremblay v. OpenAI, Inc., No. 23-cv-03223, 2024 WL 3748003 (N.D. Cal. 2024):
Plaintiffs filed a class action for copyright infringement against OpenAI, alleging their copyrighted books were used to train ChatGPT and that, when prompted, ChatGPT generated accurate summaries of their works. In order to prove their claims, plaintiffs attached an exhibit showing prompts asking ChatGPT to summarize their books as well as the outputs produced. OpenAI served a document request seeking to obtain all of the prompts and outputs plaintiffs used (including those that do not support their claims). Plaintiffs refused to do so, asserting work-product protection.
The court held that the negative test results (i.e., those which did not support plaintiffs’ claims) constituted an opinion work product as the prompts were crafted by counsel and thereby contained counsel’s mental impressions on how to interrogate ChatGPT. Therefore, because defendants could not show that counsel’s mental impressions were at issue nor that they had a compelling need for the materials, the court denied defendant’s request to compel production. However, all the prompts and outputs used for the positive testing results referenced in their complaint did have to be disclosed. The case was transferred to the Southern District of New York as part of a multi-district litigation. Currently, the case is in the discovery stage.
