Sarah Andersen, et al. v. Stability AI Ltd., et al.

• Court: District Court, N.D. California
• Decided: May 7, 2026
• Docket: 3:23-cv-00201
• Status: Unknown

Opinion

1 2 3 4 UNITED STATES DISTRICT COURT 5 NORTHERN DISTRICT OF CALIFORNIA 6 7 SARAH ANDERSEN, et al., Case No. 23-cv-00201-WHO (LJC)

8 Plaintiffs, ORDER RESOLVING DISCOVERY 9 v. DISPUTES BETWEEN PLAINTIFFS AND STABILITY 10 STABILITY AI LTD., et al., Re: Dkt. Nos. 527, 529 Defendants. 11

12 13 Plaintiffs and Defendants Stability AI, Inc. and Stability AI, Ltd. (collectively, Stability) 14 filed two discovery letter briefs. ECF Nos. 527, 529. The first addresses whether Stability must 15 add former employee William Cusick as a document custodian. ECF No. 527. The second 16 addresses the appropriate timeframe for discovery. Having considered the parties’ arguments, the 17 record in this case, and the relevant legal authorities, the Court GRANTS Plaintiffs’ request that 18 Mr. Cusick be added as an additional custodian and directs the parties to meet and confer to decide 19 on appropriate search terms and categories of custodial files to target potentially relevant 20 documents. The Court GRANTS IN PART and DENIES IN PART Plaintiffs’ request regarding 21 the appropriate timeframe for discovery. The timeframe for discovery shall start on July 1, 2021. 22 The Court assumes the parties’ familiarity with the case’s factual and procedural history 23 and the general legal standards for discovery. Its reasoning is as follows: 24 I. DESIGNATION OF MR. CUSICK 25 Plaintiffs represent that Mr. Cusick is Stability’s former Creative Director, and, in that role, 26 led Stability’s “marketing and artist relations for the launch of DreamStudio and Stable 27 1 Diffusion.” ECF No. 526-1 at 2.1 They contend that his custodial file will likely include 2 documents that are “directly relevant to disputed issues, such as copyright ownership and fair use.” 3 Id. Stability’s main objection—and the threshold issue in resolving this dispute—is that Plaintiffs’ 4 belated request to add Mr. Cusick did not comply with the Court’s prior order that Plaintiffs 5 request any additional custodians by January 30, 2026. Id. at 4; see ECF No. 392. 6 The Court’s minute order at ECF No. 392 required Plaintiffs to “send all requests for 7 additional custodians to the respective Defendants no later than January 30, 2026,” and set 8 February 19, 20262 as the deadline for parties to file letter briefs regarding additional custodians. 9 Plaintiffs requested that Stability add three additional custodians (exclusive of Mr. Cusick) on 10 January 15, 2026. See ECF No. 527-1 (Kamdar Decl.) ¶ 3. Stability responded to this request on 11 February 4, 2026 and February 21, 2026, and the parties met and conferred regarding the three 12 additional custodians on March 9, 2026. Id. ¶ 4. Per Stability’s counsel, “Plaintiffs represented 13 that they did not anticipate requesting additional custodian” and Stability subsequently “agreed to 14 add the three custodians requested by Plaintiffs ‘on the basis of your representation on the call that 15 you do not anticipate adding any further custodians.’” Id. Plaintiffs do not dispute this summary 16 of the meet and confer process. See ECF No. 526-1 at 2-4. Plaintiffs asked Stability to designate 17 Mr. Cusick as an additional custodian on March 31, 2026, two months after the Court’s January 18 30, 2026 deadline and approximately three weeks after Stability agreed to the other three 19 custodians. See Kamdar Decl. ¶ 5. Plaintiffs contend that they did not request Mr. Cusick until 20 the end of March because documents indicating that he likely had relevant information were 21 produced in late December, January and February, and that it took them additional time to 22 evaluate Stability’s production, review publicly available information regarding Mr. Cusick’s role 23 at Stability, and look “for overlap with Stability’s agreed custodians.” ECF No. 526-1 at 3. 24 A case schedule, including discovery deadlines, “may be modified only for good cause.” 25 Fed. R. Civ. P. 16(b)(4). “Although the existence or degree of prejudice to the party opposing the 26

27 1 A redacted copy of the letter brief is filed on the public docket at ECF No. 527. An unredacted 1 modification might supply additional reasons to deny a motion, the focus of the inquiry is upon the 2 moving party’s reasons for seeking modification.” Johnson v. Mammoth Recreations, Inc., 975 3 F.2d 604, 609 (9th Cir. 1992). “If that party was not diligent, the inquiry should end.” Id. But 4 “reasonable diligence does not demand perfection.” M.H.C. v. Cnty. of Los Angeles, No. CV 18- 5 8305, 2019 WL 13242561, at *4 (C.D. Cal. Nov. 5, 2019) (internal quotations omitted). As this 6 Court previously explained, “diligence is not a binary question,” and that a “more diligent attorney 7 might have” behaved differently is not dispositive. Deckers Outdoor Corp. v. Last Brand, Inc., 8 No. 23-cv-04850, 2025 WL 351596, at *2 (N.D. Cal. Jan. 31, 2025). Moreover, district courts 9 have “broad discretion in supervising the pretrial phase of litigation, and its decisions regarding 10 the preclusive effect of a pretrial order” are covered by the scope of that discretion. C.F. ex rel. 11 Farnan v. Capistrano Unified Sch. Dist., 654 F.3d 975, 984 (9th Cir. 2011). 12 The Court notes that given that the majority of documents showing Mr. Cusick’s alleged 13 involvement with issues important to this case were not produced until January and February, 14 Plaintiffs would have been hard-pressed to identify him by the January 30 deadline. While it 15 certainly would have been preferable for Plaintiffs to have identified Mr. Cusick before the March 16 9 meet and confer, Plaintiffs’ explanation that it took time to review recently-produced documents, 17 compare them to other custodians’ documents, and research Mr. Cusick’s role at Stability is 18 reasonable. Diligence does not demand perfection. M.H.C., 2019 WL 13242561, at *4. The 19 Court concludes that the fact that it took Plaintiffs roughly two months to review documents and 20 verify Mr. Cusick’s role reflects the demands of discovery in complex litigation that involves 21 multiple defendants; in this context Plaintiffs have demonstrated sufficient diligence.3 22 Although neither party squarely addressed the risk of prejudice to Stability (see Mammoth, 23 975 F.2d at 609), on the existing record no significant prejudice would result, considering the 24 25 3 The Court previously noted that allowing parties to “avoid their obligations” under agreements reached during negotiations would “undermine the purpose of the meet and confer process.” ECF 26 No. 403. The Court’s current ruling does not contradict this principle. Per Stability, Plaintiffs represented on March 9 that they did not anticipate adding any additional custodians. Plaintiffs 27 requested Mr. Cusick after additional document review and research. Nothing before the Court 1 following. First, the Stipulated ESI Order allows parties to request up to twelve custodians, and 2 Mr. Cusick would be the eleventh custodian. See ECF No. 275 § 5(a). Second, fact discovery was 3 recently extended. See ECF No. 499. And finally, the collection and product of Mr. Cusick’s 4 custodial records will be cabined to target those sources of information that are most likely to 5 contain relevant information. As directed below, Plaintiffs and Stability shall meet and confer to 6 determine the appropriate scope for the search of Mr. Cusick’s custodial records. Accordingly, the 7 Court finds that Stability will not be unduly prejudiced by the addition of Mr. Cusick as a 8 custodian. 9 Plaintiffs argue that, in addition to acting diligently in requesting Mr. Cusick, Mr. Cusick 10 will likely possess relevant documents.4 In support, they quote a document reflecting that Mr. 11 Cusick participated in workshopping communications with a journalist regarding whether Stable 12 Diffusion was trained on copyrighted material. ECF No. 526-1 at 3. This document demonstrates 13 that Mr. Cusick was knowledgeable about issues central to Plaintiffs’ claims against Stability— 14 whether Stability’s models were knowingly trained on copyrighted material—and that he was 15 positioned within the company to understand the nature of the products at issue, aspects of its 16 development, the extent of concerns about copyright and potential responses to those issues. See 17 ECF No. 375-4 (Third Am. Compl.) (TAC) ¶¶ 220-25. Stability argues that this document has 18 little probative value, as Mr. Cusick’s positions were never sent to the journalist and his proposals 19 were “subject to multiple levels of review.” Id. at 5. That the proposed responses were never sent 20 and were subject to review does not disprove that Mr. Cusick’s was knowledgeable about this 21 undisputably relevant topic or that his statements were false. Plaintiffs further argue that given 22 Mr. Cusick’s marketing role, he will likely have documents that bear on fair use. Id. at 2-3. The 23 Court agrees. Fair use, among other things, considers whether and to what extent the use of 24 copyrighted material is “commercial” in nature. Campbell v. Acuff-Rose Music, Inc., 510 U.S. 25 569, 578 (1994) (quoting 17 U.S.C. § 107(1)). It is likely that given Mr. Cusick’s role launching 26 4 To the extent Plaintiffs are arguing that they are entitled to add Mr. Cusick as a matter of right 27 under the Stipulated ESI Order (ECF No. 275) and the Court may not entertain Stability’s 1 and marketing Stable Diffusion, he will have documents that bear on the commercial nature of 2 Stability’s products. Moreover, given Mr. Cusick’s position leading “artist relations,” Plaintiffs’ 3 argument that his custodial file may include communications or internal documents regarding 4 artists’ concerns regarding copyright infringement and risk of market substitution is reasonable. 5 The Court accordingly concludes that Plaintiffs’ have shown Mr. Cusick’s custodial files will 6 likely have relevant documents. 7 Stability’s argument that documents showing Mr. Cusick’s alleged involvement include 8 other document custodians is not a sufficient basis to deny Plaintiffs’ request. That some 9 responsive documents “will necessarily be found in other custodians’ records is not sufficient to 10 defeat a search of” a custodian’s files. Shenwick v. Twitter, Inc., No. 16-cv-05314, 2018 WL 11 833085, at *1 (N.D. Cal. Feb. 7, 2018). Moreover, the documents Plaintiffs cite to were obtained 12 from other custodians’ files. It is thus unremarkable—and not meaningfully probative of 13 unreasonable cumulativeness—that they include other custodians in addition to Mr. Cusick. 14 Plaintiffs argue that Mr. Cusick, unlike the other designated custodians, worked in product 15 marketing and artist relations. See id. at 3-4; see also, Fed. R. Civ. P. 26(b)(2)(C)(i). Plaintiffs’ 16 argument that his custodial file will likely contain unique documents regarding marketing 17 Stability’s products and interfacing with artists and other users is thus persuasive. 18 The Court accordingly GRANTS Plaintiffs’ request that Mr. Cusick be designated as an 19 additional custodian. 20 However, as the Court directed during the April 16, 2026 status conference, the parties 21 must set “limited parameters” regarding the search of Mr. Cusick’s custodial file. ECF No. 504 at 22 27-28. This may include only searching select sources of electronically stored information and 23 using tightly crafted search terms designed to hit on relevant documents. The parties shall meet 24 and confer forthwith to decide on the appropriate scope and shall file a status report advising the 25 Court of their progress by May 13, 2026. If the Court is not satisfied that the parties have met and 26 conferred in good faith, the parties may be ordered to attend an in-person meet and confer session 27 in Courtroom G on May 15, 2026. II. TIMEFRAME FOR DISCOVERY 1 In the letter brief at ECF No. 529, Plaintiffs and Stability dispute whether the discovery 2 timeframe should begin on January 1, 2021 or January 1, 2022. Although “the temporal scope of 3 discovery . . . should be limited to a reasonable time based on the parties’ allegations,” what 4 constitutes a reasonable time “must be determined on a case-by-case basis.” Garedakis v. 5 Brentwood Union Sch. Dist., No. 14-cv-04799, 2016 WL 1133715, at *3 (N.D. Cal. Mar. 23, 6 2016); In re Bofl Holding, Inc. Sec. Litig., No. 15-cv-02324, 2021 WL 1812822, at *5 (S.D. Cal. 7 May 6, 2021). “In general, courts allow discovery to extend to events before and after the period 8 of actual liability so as to provide context.” Hatamian v. Advanced Micro Devices, Inc., No. 14- 9 cv-00226, 2015 WL 7180662, at *2 (N.D. Cal. Nov. 16, 2015) (internal quotations omitted) 10 (collecting cases). 11 In general terms, Plaintiffs’ claims against Stability assert that Stability’s image generation 12 models were trained on datasets that included Plaintiffs’ registered works. See TAC ¶¶ 220-23. 13 As is relevant here, they allege that Stability’s Stable Diffusion 2.0 model, launched in November 14 2022, was trained “[b]etween April and November 2022” on the LAION-5B datasets “and subsets 15 5B,” and that “Stability knew the LAION-5B contained copyrighted works.” Id. ¶¶ 221-22. 16 LAION (“Large-Scale Artificial Intelligence Open Network”) is an organization based in 17 Germany that develops and releases large-scale machine learning models and datasets available to 18 the public. Id. ¶ 57. It released the LAION-400M dataset in August 2021 and released the larger 19 LAION-5B dataset in October 2022. Id. ¶¶ 59, 65. (LAION-400M is a subset of LAION-5B. See 20 id. ¶ 65.) Plaintiffs explain in their portion of the letter brief that “the data for LAION 5B was 21 filtered from September 2021 to January 2022.” ECF No. 528-1 at 2.5 22 Plaintiffs contend that Stability funded LAION’s creation of LAION-5B. Id. Stability 23 argues that this is overselling Stability’s role, and the interview Stability cites to in support of their 24 position suggests that Stability became involved “[b]y the end 2021” when the LAION-5B dataset 25 was nearly seventy percent complete. Id. at 4.6 While Stability may not have funded the creation 26

27 5 A redacted copy of the letter brief is filed on the public docket at ECF No. 529. The redacted 1 of LAION-5B from its very inception, Stability was involved at least by early 2022 and funded 2 LAION-5B’s completion. As Plaintiffs quite sensibly reason, it is unlikely that Stability 3 “suddenly support[ed] the creation of 5B” without having some background knowledge of 4 LAION. ECF No. 528-1 at 3. It is highly plausible that Stability conducted due diligence, 5 researched LAION’s capabilities, and held internal discussions regarding the risks and utility of 6 LAION’s datasets before deciding to fund 5B and subsequent datasets. Plaintiffs are thus entitled 7 to explore what Stability knew about LAION’s dataset development in the months leading up to 8 Stability’s decision to fund 5B, as this will likely be relevant to show Stability’s “knowledge, 9 intent, and willfulness.” Id. 10 The Court concludes that starting discovery on July 1, 2021, slightly before LAION 11 released 400M, will likely encompass Stability’s earlier communications and research regarding 12 LAION’s datasets and “provide context” regarding Plaintiffs’ claims. Hatamian, 2015 WL 13 7180662, at *2. 14 Although Stability complains that Plaintiffs delayed objecting to Stability’s proposed 15 January 1, 2022 start date, discovery may be conducted in an iterative manner. See ECF No. 528- 16 1 at 6. Subsequent requests may be reasonable after the requesting party has learned from 17 information that is gathered earlier in the discovery process. Whether such requests are permitted 18 depends on the usual considerations, such as discovery cutoffs or other case management 19 deadlines, and the proportionality and relevancy principles established in Rule 26 of the Federal 20 Rules of Civil Procedure. Plaintiffs explain that they were waiting on Stability’s discovery 21 responses to determine if they would preempt the need for discovery from 2021; they apparently 22 did not, and Plaintiffs then raised the issue with Stability. Id. at 3-4. That Plaintiffs had to expand 23 their requests in response to ongoing discovery does not show undue delay or warrant imposing a 24 January 1, 2022 start date for discovery. In any event, extending the discovery timeframe an 25 additional six months is not disproportionate to the needs of the case, given the parties’ relative 26

27 Interview with LAION co-founder Christoph Schuhmann, MLCon by devmio (June 21, 2023), 1 access to information, the importance of the issues at stake, and the parties’ resources. See Fed. R. 2 |} Civ. P. 26(b)(1). 3 The start date for discovery is accordingly set at July 1, 2021. 4 || Il. CONCLUSION 5 As explained above, Plaintiffs’ request that Mr. Cusick be added as a document custodian 6 || is GRANTED. The parties are directed to meet and confer regarding the scope of the collection 7 || and production of Mr. Cusick’s custodial records, and file a status report no later than May 13, 8 || 2026. Plaintiffs’ request regarding the appropriate timeframe for discovery is GRANTED IN 9 || PART and DENIED IN PART. The timeframe for discovery shall start on July 1, 2021. 10 1] IT IS SO ORDERED. 13 Dated: May 7, 2026 14 Lae. | he 2 x Zl. Os = 16 Ligh Magistrate Judge (17

Z 18 19 20 21 22 23 24 25 26 27 28