In a case that has provided copious amounts of eDiscovery fodder, the United States District Court for the Northern District of California reaffirmed that prompts and related settings used by Concord Music Group’s investigations of Anthropic’s AI system are protected attorney work product, specifically opinion work product, because they reflect counsel’s mental impressions and litigation strategy. As a baseline rule, such prompts and outputs are “virtually undiscoverable” under Federal Rule of Civil Procedure 26. However, the court emphasized that work-product protection is not absolute and may be waived when a party places the substance of its investigation at issue, triggering the fairness principle that prevents a party from using privilege as both a sword and a shield.
Applying that principle, the court found that the publishers waived protection over certain generative-AI prompts and outputs from their post-suit investigation into Anthropic’s guardrails. Because the publishers intend to present testimony that their investigators were able to elicit infringing outputs using simple prompts that “anyone could do,” Anthropic has to be permitted to test and cross-examine those claims. As a result, the court ordered production of all prompts and outputs, both successful and unsuccessful, used by investigators and client witnesses in the post-suit testing. The waiver did not extend, however, to prompts crafted by attorneys that were never used by those witnesses.
By contrast, the court declined to order production of prompts and outputs from the pre-suit investigation, concluding that Anthropic’s discovery needs could be satisfied through statistical data rather than disclosure of counsel’s mental impressions. To balance fairness with privilege, the court clarified and expanded its prior order to require detailed quantitative breakdowns, “the denominator”, showing how often prompts produced infringing, disputed, or non-infringing results across pre-suit, post-suit, and large-scale sampling efforts. This approach preserves core work-product protection while ensuring transparency where generative-AI testing is used to support litigation claims.
If your organization is seeking support with eDiscovery, our team has solutions to address all phases of the discovery process. At CODISCOVR, we deliver client-focused, defensible, and scalable solutions using advanced technology and intelligent review practices to meet eDiscovery, document review, and information governance needs in a manner that reduces the risks and costs associated with electronically stored information (ESI). Reach out to Nicole Gill, Chair, Managing Member CODISCOVR. With almost a decade of experience, she manages complex and high-profile eDiscovery projects and routinely navigates data and privacy protection laws across many domestic and foreign jurisdictions.
