In re Mosaic LLM Litigation

Background

Databricks acquired MosaicML in July 2023. Plaintiffs assert a direct copyright-infringement claim against MosaicML based on the alleged use of their works to train MosaicML’s generative artificial-intelligence models. They also assert vicarious copyright infringement against Databricks, alleging that Databricks had the right and ability to control MosaicML’s alleged infringement and received a financial benefit from it.

The court had previously allowed plaintiffs to designate Dr. Ali Ghodsi, Databricks’ CEO, as a document custodian because plaintiffs showed that he was a principal decision-maker in Databricks’ acquisition of MosaicML. Since then, plaintiffs’ claims against Databricks had been narrowed to vicarious copyright infringement, and plaintiffs had reviewed documents produced by Ghodsi. Plaintiffs then sought to depose him.

Legal standard

Federal Rule of Civil Procedure 30 generally allows a party to depose any person, but Rule 26(c) permits a court, for good cause, to protect a person from annoyance, oppression, undue burden, or expense by prohibiting or limiting discovery. The party seeking protection must show the harm or prejudice that would result from the discovery.

The court discussed the heightened concern about depositions of high-level executives, sometimes called “apex” witnesses, when they have little personal knowledge of the case. It declined to treat the “apex doctrine” as a rigid set of rules. Instead, it treated the issue as an application of Rule 26’s requirements of relevance and proportionality and its protection against abusive or harassing discovery. Courts generally consider whether the executive has unique, firsthand, non-repetitive knowledge and whether the requesting party has exhausted less intrusive discovery methods.

Analysis

Plaintiffs argued that Ghodsi’s documents showed he had unique and relevant knowledge about Databricks’ alleged vicarious infringement. The court disagreed with plaintiffs’ showing. The exhibits showed that Ghodsi was involved in acquiring MosaicML and prioritized the company because of its generative-artificial-intelligence models. But the court said that this did not show that Ghodsi directed or supervised MosaicML’s allegedly infringing activities.

The court also found that the cited communications—an announcement about a new model, meeting notes in which Ghodsi asked about a possible data source, and an encouraging but non-substantive message about a generative-artificial-intelligence tool—did not show that he personally participated in decisions about MosaicML’s use of allegedly infringing training data.

The court concluded that plaintiffs had not shown that Ghodsi had sufficient relevant information about the vicarious-infringement claim to justify his deposition at that stage. It identified other possible sources of information, including Dr. Naveen Rao, MosaicML’s CEO, who became a Databricks vice president of engineering after the acquisition. The court said Rao would presumably be able to testify about MosaicML’s incorporation into Databricks and whether Databricks controlled MosaicML’s alleged infringing activity. The court also noted that plaintiffs had not noticed depositions of other Databricks founders or senior managers whose involvement might provide more direct knowledge.

The court did not rule that plaintiffs were barred from deposing executives or senior employees generally. It stated that depositions of selected Databricks financial officers or technology officers with more relevant knowledge than Ghodsi might be warranted, and that depositions of engineers or other employees who directly worked on the models could also be appropriate.

Ruling

Magistrate Judge Lisa Cisneros denied plaintiffs’ request to depose Dr. Ghodsi without prejudice. Plaintiffs may renew the request if later discovery shows that Ghodsi has personal, relevant knowledge concerning the vicarious-infringement claim that is not unreasonably cumulative or duplicative of other witnesses’ testimony. Plaintiffs may also renew the request if defendants unreasonably object to depositions of other Databricks executives or senior employees likely to possess relevant information.