Courts are increasingly addressing whether data generated through generative artificial intelligence tools, including prompts, outputs, and activity logs, must be produced in discovery. Under FRCP 26(b)(1), parties may obtain discovery of non-privileged material that is relevant and proportional to the needs of the case, and courts have emphasized that new forms of electronically stored information are not exempt simply because they are novel. As artificial intelligence tools reshape how electronically stored information is created and stored, relevant generative artificial intelligence data is discoverable and must be treated like any other potentially relevant electronically stored information.
A key development is In re OpenAI, Inc., Copyright Infringement Litigation, where Magistrate Judge Ona Wang compelled production of millions of generative artificial intelligence logs, including user prompts and model responses, subject to anonymization of user references. No. 25-MD-3143, 2025 WL 3468036 (S.D.N.Y. Dec. 2, 2025). The court concluded these logs were relevant and proportional to plaintiffs’ claims that the defendant’s artificial intelligence systems reproduced copyrighted works in their outputs, and found that privacy concerns could be mitigated through anonymization and protective orders rather than serving as a categorical bar to production. In a separate ruling in the same litigation, Magistrate Judge Wang denied a motion to compel the New York Times to produce content from its internal artificial intelligence tools, crediting arguments that review of approximately 80,000 entries would take more than 1,300 hours and finding the request irrelevant and disproportionate. No. 25-MD-3143 (S.D.N.Y. Sept. 19, 2025).
These early rulings underscore that relevance and proportionality remain central to decisions on discoverability of generative artificial intelligence data. Organizations are advised to identify whether custodians use generative artificial intelligence tools, how they are used, and where prompts, outputs, and activity logs are stored, including on third-party platforms. When litigation is anticipated, parties should preserve generative artificial intelligence data that relates to claims or defenses by disabling auto-delete where appropriate, exporting chat histories, saving key exchanges, and coordinating with information technology teams on logs and metadata, while avoiding edits that alter context. Parties should address scope, relevance, and proportionality of generative artificial intelligence data in early electronically stored information protocols and meet-and-confer discussions, use protective orders and anonymization to manage confidentiality and privacy concerns, and update information governance, legal hold procedures, and retention policies to account for generative artificial intelligence data. As generative artificial intelligence data increasingly sits at the heart of disputes, courts are not carving out exemptions, but proportionality remains a meaningful limit that can constrain overly burdensome or tangential requests.
