In re Mosaic LLM Litigation

Background

The plaintiffs filed a discovery letter brief disputing whether they could take five third-party depositions: one of Amazon Web Services, one of former MosaicML employee Nicholas Sapp, and three of the defendants’ customers. The court had previously allowed five additional party depositions and extended the fact-discovery deadline to December 5, 2025. The plaintiffs also sought to extend the discovery cutoff to January 5, 2026, but Judge Breyer had not ruled on that request when this order was issued.

The operative complaint asserted a direct copyright-infringement claim against MosaicML and a claim for vicarious copyright infringement against Databricks. The plaintiffs had also asked to amend the complaint to revive a direct-infringement claim against Databricks and add claims for contributory infringement and inducement of infringement. The court noted that the proposed third-party discovery appeared more obviously related to those proposed claims, but it also found that the discovery could relate to the existing claims and the defendants’ fair-use defense.

Legal standard

The court applied Federal Rule of Civil Procedure 26, which generally limits discovery to nonprivileged information relevant to a claim or defense and proportional to the needs of the case. In deciding whether to allow additional depositions, the court considered factors including the importance of the issues, the burden and expense of the discovery, whether the request was cumulative or duplicative, whether the information could be obtained from a more convenient or less burdensome source, and whether the discovery could be completed within the existing schedule.

Analysis

The court found that the proposed depositions were relevant because information about how the defendants used and distributed the plaintiffs’ works could bear on the fair-use defense. The proposed witnesses included a platform allegedly used to host and share datasets, customers to whom the defendants allegedly distributed datasets, and a former employee allegedly familiar with the distribution and the procedures for making datasets available to customers.

Even so, the court held that allowing all five depositions was not proportional to the needs of the case. The plaintiffs waited until seven days before the close of fact discovery to notify the defendants of their intent to seek the depositions. The court found that the plaintiffs knew about the dataset-distribution issue earlier and should have pursued third-party depositions sooner. Allowing the depositions at that stage would disrupt the case schedule.

The court also found that the proposed depositions would partly duplicate discovery already obtained from the defendants, including witness testimony and documents concerning alleged sharing of datasets with enterprise customers. The court further noted that the plaintiffs were already seeking AWS access logs, making a deposition of AWS less necessary because some information could be obtained through a less burdensome process.

The court therefore denied the request to depose AWS, Nicholas Sapp, and the three identified customers without prejudice. It stated that the plaintiffs could renew the request if Judge Breyer changed the case schedule or allowed the complaint to be amended, or if ongoing discovery produced information that significantly changed the relevance and proportionality analysis.

Hanlin Tang deposition

The plaintiffs also argued that Hanlin Tang, the defendants’ organizational witness under Rule 30(b)(6), had not been prepared to answer fully about Topics 37 and 40. Those topics concerned providing access to training data to customers or the public and technological support for downloading and using datasets to train large language models.

The court explained that an organization must make a good-faith effort to designate and prepare a knowledgeable witness, using information reasonably available to it, including information from former employees. Tang testified that Nicholas Sapp was likely knowledgeable about how customers received access to the datasets, and the defendants’ counsel confirmed that Sapp had not been contacted before Tang’s deposition. The court concluded that this supported reopening Tang’s deposition for limited questioning on Topics 37 and 40.

Disposition

The court denied the plaintiffs’ request for five third-party depositions without prejudice. It ordered that Tang’s Rule 30(b)(6) deposition be reopened for two additional hours on Topics 37 and 40, required the defendants to prepare him as required by Rule 30(b)(6), and granted the parties’ joint administrative motion to file excess-page exhibits.