In re Mosaic LLM Litigation

4 UNITED STATES DISTRICT COURT 5 NORTHERN DISTRICT OF CALIFORNIA 7 Case No. 24-cv-01451-CRB (LJC) In Re Mosaic LLM Litigation ORDER RESOLVING DISCOVERY 9 LETTER BRIEF AT ECF NO. 127 10 Re: Dkt. No. 127

14 Pending before the Court is the parties’ Joint Discovery Letter regarding whether the CEO of Defendant Databricks, Dr. Ali Ghodsi, may be designated as a document custodian. ECF No. 127. Plaintiffs argue that Ghodsi is “uniquely situated to provide relevant information[,]” given his leadership role and involvement in the acquisition of Defendant MosaicML. Id. at 1.

18 Defendants disagree, characterizing Plaintiffs’ arguments as mere “speculation that [Ghodsi] must have unique information simply by virtue of serving as CEO.” Id. at 5. For the following reasons, the Court GRANTS Plaintiffs’ request to designate Ghodsi as a document custodian.

21 Defendants frame their objections to Ghodsi serving as a custodian as being “on the basis of Apex considerations.” Id. at 4. The apex doctrine “is essentially the courts’ recognition that deposing high-level officers of large corporate and governmental entities imposes meaningful burdens on those entities and witnesses, in part because in the absence of any restrictions, sitting for depositions could quickly become a full-time job for leaders of entities that face extensive litigation.” In re Uber Techs., Inc., Passenger Sexual Assault Litig., No. 23-md-03084-CRB (LJC), 2025 WL 896412, at *1 (N.D. Cal. Mar. 24, 2025). The Court notes that the apex doctrine Co., Ltd, 282 F.R.D. 259, 263 (N.D. Cal. 2012) (analyzing the apex doctrine as applying to depositions of high-level executives); In re Facebook, Inc. Consumer Priv. User Profile Litig., No. 3:18-MD-02843, 2021 WL 10282213, at *10 (N.D. Cal. Nov. 14, 2021) (finding that the apex doctrine did not apply to a dispute regarding whether executives “should be added as document custodians”). However, the parties’ stipulated ESI Protocol expressly provides that, if the requesting party (here, Plaintiffs) requests to designate three additional document custodians, the producing party may “object” to Plaintiffs’ selections “only on the basis of Apex considerations[.]” ECF No. 61 at 2. The Court accordingly considers the appropriate apex principles—namely, the executive’s access to relevant information, such information’s uniqueness, and burden of production—in this decision. See Apple v. Samsung, 282 F.R.D. at 263 (rejecting the “two-prong test” of permitting apex depositions only if the deponent has “unique first-hand, non-repetitive knowledge” and other “less intrusive discovery methods” have been exhausted in favor of a “more nuanced equation” considering relevance, burden, unique knowledge, and availability of other discovery methods).

15 Plaintiffs argue that Ghodsi “near undeniably has unique documents relevant to Plaintiffs’ vicarious infringement claim,” as he was the “architect, the engine, and the closer” of Databricks’ acquisition of MosaicML. ECF No. 127 at 2. In support of this contention, they cite to an interview given by Ghodsi where he discussed meeting MosaicML’s CEO at a conference and deciding to try to purchase MosaicML based on his believe that MosaicML’s generative AI models were extremely valuable. Id. The Court finds it credible that, given Ghodsi’s role in the acquisition, he will have documents regarding Databricks’ acquisition of Mosaic, which are arguably relevant to Databricks’ knowledge of Mosaic’s alleged infringing activity and the financial benefit Databricks believed it could obtain from Mosaic’s infringing activity. See In re Glumetza Antitrust Litig., No. 19-cv-05822, 2020 WL 3498067, at *7 (N.D. Cal. June 29, 2020).

25 Defendants try to distinguish this case from Kadrey v. Meta, where the plaintiffs sought to “add Meta’s CEO based on produced documents indicating that he was a principal decision-maker on acquiring the datasets at issue.” ECF No. 127 at 5. Plaintiffs cite to interviews where Ghodsi datasets including Plaintiffs’ copyrighted works. As in Kadrey v. Meta, Ghodsi’s role as a “principal decision-maker” supports Plaintiffs’ position that he has relevant information. While it is somewhat more speculative that Ghodsi has documents regarding Databricks and MosaicML’s relationship post-acquisition, given Ghodsi’s role in acquiring MosaicML and integrating it into Databricks, it seems highly plausible that Ghodsi has information regarding their ongoing relationship.1 Judge Breyer’s recent decision to allow Plaintiffs to add a direct infringement claim against Databricks strengthens Plaintiffs argument that Ghodsi will have relevant documents and is an appropriate custodian. See ECF No. 129.

9 Defendants rely heavily on the argument that Plaintiffs have not shown Ghodsi will have unique documents, contending that he does not have any documents relevant to Plaintiffs’ claims that “would not be in corporate documents, [former Mosaic CEO] Dr. Rao’s documents, or documents from the other eleven custodians.” ECF No. 127 at 4. But that there will be overlap between Ghodsi’s custodial documents and the other custodians’ is not dispositive. “The idea that responsive documents will necessarily be found in other custodians’ records is not sufficient to defeat a search of [a custodian’s] files.” Shenwick v. Twitter, No. 16-cv-05314, 2018 WL 833085, at *1 (N.D. Cal. Feb. 7, 2018); see In re Uber Techs., 2025 WL 896412, at *2 (explaining, in the context of depositions, that “an overemphasis on unique knowledge is inconsistent with otherwise Defendants argue that Plaintiffs claim that Ghodsi has unique and relevant information is based on speculation. See ECF No. 127 at 5. Certainly, a requesting party with limited access into a responding party’s decision-making process and corporate structure often must engage in some degree of speculation when choosing custodians. Given this information imbalance, the responding party is generally best positioned to select custodians, as they know and can “identify those individuals within its organization likely to have information relevant to the case.” In re EpiPen (Epinephrine Injection, USP) Mktg., Sales Pracs. & Antitrust Litig., No. 17-md-2785, 2018 WL 1440923, at *2 (D. Kan. Mar. 15, 2018); see The Sedona Principles, Third Edition, 19 Sedona Conf. J. 1, 118 (2018) (explaining, in the introduction to Principle 6, that “a responding party is best situated to preserve, search, and produce its own ESI”). The ESI Protocol reflects this general principle, as it allows the responding party to select their first ten document custodians. ECF No. 61 at 2. Once the responding party has identified ten custodians, the requesting party can identify three more, subject to limited objections. Id. Plaintiffs have followed this protocol, and, based on the limited information available to them, made a selection they believe will produce relevant information. While their decision to select Ghodsi inevitably involved some degree of speculation, they “will have to bear the consequences” of making an unwise selection. Tremblay v. OpenAI, Inc., 23-cv-03223 (N.D. Cal. Ot. 18, 2024), ECF No. 191 at 3.

1 common approaches to discovery and trial advocacy, where corroborating evidence and testing || credibility through the testimony of more than one witness is often appropriate”). Moreover, Ghodsi has a unique leadership role relative to other custodians. The majority of agreed-upon custodians, identified at ECF No. 127-1, are members and directors of MosaicML’s research and engineering teams. As Plaintiffs argue, Ghodsi, as CEO of the parent company Databricks, may || well have unique information regarding “strategic-level LLM development commercialization, including communications with Databricks’ board and investors.” ECF No. 127 at 2.

8 Defendants do not argue that designating Ghodsi as a document custodian would be unduly || burdensome. As the parties agreed to initially identify up to thirteen document custodians, and || Defendants do not contend that it would be uniquely burdensome to produce Ghodsi’s custodial 11 file, the Court has no reason to find that designating Ghodsi as one of the thirteen custodians || would pose an undue burden on Defendants.

5 13 Plaintiff's request to designated Ghodsi as a document custodian is accordingly |] GRANTED.

3 15 IT IS SO ORDERED.

16 Dated: June 26, 2025 1g 7 LIBA 4. CISNEROS 19 Liga CIs Magistrate Judge