What You Need to Know in a Minute or Less

Artificial intelligence tools are rapidly reshaping how ESI is created and stored, particularly with respect to content generated by large language models. As companies adopt generative AI (GenAI) tools for drafting, summarizing, analyzing, and other business uses, courts are grappling with whether GenAI Data such as prompts (what a user types),outputs (what the AI tool generates), and activity logs (data about when and how tools were used) fall within traditional discovery obligations. The details are evolving, but recent court decisions make two things clear:

1. Relevant GenAI Data is discoverable; and
2. Parties must treat it like any other potentially relevant ESI

Traditional Discovery Rules Still Govern Non-Traditional Data

Under FRCP 26(b)(1), parties may obtain discovery of non-privileged material that is relevant and proportional to the needs of the case. Courts have made clear that new forms of ESI are not exempt simply because they are novel. Traditional discovery principles apply equally to emerging sources of ESI, including GenAI Data.

Key Early Decisions on GenAI Data Discoverability

The most defining ruling so far as to GenAI Data discoverability is In re OpenAI, Inc., Copyright Infringement Litigation, where Magistrate Judge Ona Wang compelled production of millions of GenAI logs, including user prompts and model responses, on the condition that user references be anonymized. 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 AI systems reproduced copyrighted works in their outputs. The decision emphasized that privacy concerns can be mitigated through anonymization and protective orders and do not categorically bar production of AI output. 

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 AI tools, finding the request both irrelevant and disproportionate. The New York Times argued that review of approximately 80,000 entries would take more than 1,300 hours—a substantial burden given the data’s limited connection to the issues. No. 25-MD-3143 (S.D.N.Y. Sept. 19, 2025).

Relevance and Proportionality Still Reign

These rulings underscore two key discovery concepts:

1. Relevance: GenAI Data is discoverable when tied to a claim or defense.
2. Proportionality: Even massive volumes of GenAI Data may be discoverable when justified by the needs of the case, but proportionality remains a highly relevant inquiry.

GenAI and E-Discovery in Practice

Given the rapidly evolving role of GenAI in all aspects of daily life, parties must be well-prepared to address it head-on in discovery. Since it is rarely reasonable or proportional to preserve all GenAI Data, developing a defensible approach that is targeted, reasoned, and well-documented is critical at the early stages of the engagement.

Identify Relevant GenAI Data

Determine if any custodians of potentially relevant data use GenAI tools, how the tools are used, and where prompts and outputs are stored. Keep in mind that relevant activity logs may exist separately, including on third-party platforms.

Preserve What’s Potentially Relevant

When litigation is anticipated, preserve GenAI Data that relates to claims or defenses, particularly where the GenAI Data may contain factual assertions or substantive content. Steps vary by platform but may include disabling auto-delete settings, exporting chat histories, saving key exchanges in document repositories, and coordinating with IT to understand retention of logs and metadata. Custodians should not edit or selectively copy GenAI Data in ways that alter context and should disclose use of personal or browser-based tools so those sources can be evaluated. Specific preservation measures will depend on the matter and the systems at use; litigators should be prepared to oversee preservation efforts and provide instructions to custodians and client IT during the legal hold process.

Negotiate Scope Early

If GenAI Data is implicated, address relevance and proportionality in ESI protocols and early meet-and-confer discussions. Clear definitions and targeted limits can prevent fishing expeditions and reduce cost and burden.

Address Confidentiality

Take privacy concerns seriously. Where possible, use protective orders and anonymization protocols to manage sensitive information while meeting discovery obligations.

Update Information Governance

Incorporate GenAI Data into ESI inventories, legal hold procedures, and retention policies to improve discovery readiness. AI-specific policies surrounding acceptable use and data confidentiality also should be considered.

Conclusion

GenAI Data discoverability is quickly becoming a central issue in e-discovery. Courts are not carving out exemptions for GenAI Data; traditional discovery principles still apply. When GenAI Data goes to the heart of a dispute, it likely will be discoverable, but proportionality remains a meaningful limit. Companies and their litigation teams should address GenAI Data early in discovery planning, work closely with e-discovery specialists to minimize burden, and proactively manage privacy concerns.

Be sure to contact the firm’s E-Discovery Analysis and Technology (e-DAT) team early to ensure GenAI discovery issues are anticipated and managed strategically—avoiding disputes, minimizing disruption and expense, and aligning discovery with case goals.

Stay tuned for an upcoming Litigation Minute on the intersection of GenAI Data with the attorney-client privilege and work product doctrine.