In re Uber Techs., Inc. Passenger Sexual Assault Litig.
Citation:2025 WL 2816602 (N.D. Cal. 2025)
Summary:The court ordered Uber to produce data from its Flack database related to sexual assault incidents from 2012 to 2024 and to provide a corporate witness for deposition. The court emphasized the importance of this data for accurately counting incidents, while also addressing issues with the S-RAD system, ultimately ordering further deposition but denying additional requests as unduly burdensome. The rulings underscore the significance of ESI in litigation.
Court:United States District Court, N.D. California
Date decided:
Judge:Cisneros, Lisa J.
IN RE: UBER TECHNOLOGIES, INC., PASSENGER SEXUAL ASSAULT LITIGATION This Document Relates To: ALL CASES Case No. 23-md-03084-CRB (LJC) United States District Court, N.D. California Filed October 03, 2025 Cisneros, Lisa J., United States Magistrate Judge ORDER RESOLVING DISCOVERY LETTERS REGARDING FLACK AND S-RAD *1 This Order addresses the parties' second joint discovery letter regarding data from Uber's Flack database (Dkt. No. 3914) and their joint discovery letter regarding Uber's S-RAD system for assessing passenger risk (Dkt. No. 4009). The Court presumes the parties' familiarity with the record, the history of the case, and applicable law. A. Second Joint Letter Regarding Flack Data The parties have filed a second joint letter addressing Flack, a system that Uber uses to consolidate and analyze reports submitted through Bliss and Jira, other programs that have already been subject to discovery. The Court previously determined that Plaintiffs' requests for discovery related to Flack were not barred by the deadline for substantial completion of corporate discovery common the MDL, and ordered as follows: Uber shall produce no later than September 9, 2025, either: (1) a declaration confirming that the incident numbers that Uber produced in response the interrogatories at issue are the same numbers that would have been displayed in Flack and/or the Dashboard; or (2) if the numbers are different, the numbers as they existed in those systems as of the dates that Uber produced its previous responses. If Plaintiffs believe additional discovery is warranted, then by September 12, 2025, the parties shall meet and confer regarding a limited set of discovery requests, seeking additional data or documents from Flack or the Dashboard. If the parties are unable to reach an agreement, they shall file a Joint Discovery Letter by September 15, 2025. Dkt. No. 3848 at 5 (footnoted omitted). Plaintiffs now seek: (1) an order requiring Uber to produce “all data in Flack related to sexual assault or sexual misconduct ... incidents between January 1, 2012 through December 1, 2024, including from all fields identified in” an exhibit to the joint letter; (2) “a corporate witness, no later than October 3, 2025, for a Rule 30(b)(6) deposition up to four hours on Uber's use of Flack and the process by which Uber identified, exported, and produced the Flack data”; and (3) a briefing schedule for a motion for preclusive sanctions under Rule 37 of the Federal Rules of Civil Procedure. Dkt. No. 3914 at 3–5. Uber contends that Plaintiffs' requests are procedurally improper because they do not correspond to a Rule 34 document request and that they are untimely. Id. at 5–7. Uber does not offer any counterproposal for more limited discovery. Throughout this litigation, Plaintiffs have sought discovery of the number and nature of incidents of sexual assault reported to Uber. Uber repeatedly represented that Bliss and Jira were the best (and perhaps only) sources of such reports, and that it could not provide numbers of incidents because it sometimes received or created multiple reports related to the same incident. As it turned out, however, Flack—which Uber did not disclose as a non-custodial source of relevant information—consolidates reports into discrete incidents, and allows Uber to count the number of reported incidents of particular categories of sexual misconduct. *2 That consolidation process relies on an automated logic system that may not be perfect, but Hannah Nilles, Uber's Head of Safety for the Americas, testified at her July 23, 2025 deposition that she uses numbers of incidents derived from Flack and considers them reliable: Q. Okay. So you go into the data, say okay, what was the incident rate for critical sexual assault? A. Yes. Q. And you consider that information in doing your job? A. Yes. .... Q. Okay. So the number that you're looking at when you're making those decisions, do you feel comfortable relying on that number in making your decisions? .... A. Yes, I do. .... Q. Where is that data housed? A. In the database. Q. What is the database called? A. Flack. .... Q. Okay. So as the head of safety for the Americas, that data about incident rate that you get from flack [sic], that's data that you -- you rely on in doing your job, right? A. Yes. Q. Okay. And you consider it to be accurate? .... A. If you're speaking specifically to the incident rates that I look at, yes. Dkt. No. 3993-23 (provisionally under seal) at 4 (Nilles Dep. at 66:4–67:19, 68:17–69:3). There is other prior evidence in the record that also addressed, at least to some degree, Flack's function and relevance. At Uber witness Katy McDonald's April 24, 2025 deposition, for example, Plaintiffs' counsel questioned her about a document indicating that “Flack uses logic to combine multiple tickets on the same trip UUID to a single event.” Dkt. No. 3993-21 (provisionally under seal) at 3 (McDonald Dep. at 45:14–15). But Plaintiffs assert that Uber repeatedly downplayed Flack's reliability and the extent to which it includes data beyond what Uber had already produced from Bliss and Jira. That Plaintiffs prepared, but ultimately did not pursue, a discovery letter regarding Flack dated May 1, 2025 lends credibility to Plaintiffs' contentions that Uber persuaded them that Flack did not have further relevant information. To the extent there is any question, the Court is inclined to give Plaintiff's the benefit of the doubt on this issue in light of Uber's repeated failure to identify Flack as a relevant source of data earlier in the case. As but one example, a meet-and-confer letter that Uber sent Plaintiffs in February 2024 failed to include Flack in list of “custodial and non-custodial sources relevant to this litigation,” including in a subset of that list for sources of “Structured Data” that included Bliss, Jira, and other databases. Dkt. No. 3993-10 (provisionally under seal) at 9–10. If Uber had been forthcoming, this issue could have been resolved well over a year ago. Accordingly, the Court stands by its previous conclusion that Plaintiffs' requests for Flack data are not barred by the June 16, 2025 substantial completion deadline, which occurred well before Hannah Nilles's deposition. Plaintiffs' renewed interest in Flack data logically followed from Nilles's testimony, and fact discovery had not closed as of the date that Plaintiffs raised the issue. As for whether Plaintiffs' current request falls within the scope of any existing request for production of documents under Rule 34 of the Federal Rules of Civil Procedure, the Court tends to agree with Uber that none of the requests Plaintiffs have invoked is a clear fit for the data they now seek. Request No. 73 seeks documents regarding “the total number of reports or claims of SEXUAL MISCONDUCT or SEXUAL ASSAULT by DRIVERS on RIDERS.” Dkt. No. 3913-4 (provisionally under seal) at 61 (emphasis added). The data Plaintiffs now seek relates to individual incidents, rather than the total number of reports. Request No. 83 seeks documents related to Uber's Safety Reports, id. at 68, but Plaintiffs have not explained how this data falls within that request. The closest, perhaps, is request No. 84, seeking “[a]ny and all DOCUMENTS RELATED TO [Uber's] reviews, studies, surveys, or risk or hazard assessments related to SEXUAL MISCONDUCT or SEXUAL ASSAULT of RIDERS,” id. at 69, but to construe that request as seeking all data of any kind that Uber had regarding any reported incidents of sexual assault would be a stretch. *3 That said, the fault for Plaintiffs' imprecise requests appears to rest in significant part on Uber. As noted above, Uber did not disclose the existence of Flack in its disclosures of Bliss and Jira as potentially relevant non-custodial sources for discovery, and Flack therefore did not come up in the monthslong disputes about the scope of discovery from Bliss and Jira. Moreover, as far as the Court can tell, Uber represented throughout this litigation that it did not have numbers of total incidents (as opposed to tickets or reports) in various categories. In the parties' previous joint letter on this topic, Uber acknowledged that it responded to Plaintiffs' interrogatory seeking the number of incidents in various categories of sexual assault with instead the number of reports for such categories, and that “the numbers of combined ticket categorizations” available from Flack are “obviously not” “the same as the numbers for the non-combined categorizations” Uber previously provided. Dkt. No. 3700 at 7. Though Uber now offers arguments for why the Flack data would not have fully answered Plaintiffs' interrogatories, the clear implication of Uber's response was that it did not have data (or at least reliable data) categorized by incident rather than by ticket. Nilles's testimony to the contrary reasonably prompted Plaintiffs to seek that data. Finally, while Uber suggests that Plaintiffs' request improperly exceeded the “limited” further discovery contemplated by the Court's previous Order, Uber has neither suggested a more limited scope of production nor articulated any particular burden in exporting a larger set of data from the Flack database as compared to producing a subset of such data. Accordingly, the Court grants Plaintiffs leave to serve a Rule 34 request for the data sought in this joint letter no later than one business day from the date of this Order, and sets a deadline of one week after Plaintiffs' request for Uber's production. Uber shall produce a witness for a Rule 30(b)(6) deposition not to exceed four hours no later than one week after that production, unless Plaintiffs agree to a later date. B. Joint Letter Regarding S-RAD On September 24, 2025, the parties filed a joint letter in which Plaintiffs seek the following verifications and productions related to S-RAD, a machine learning model Uber has used to assess safety risks of particular pairings between drivers and riders: A written, verified confirmation that: 1) “supply plans” are deleted after 30 days; 2) deletions happen on a rolling basis; 3) this policy, to date, has never been suspended; and 4) this policy has been in effect since S-RAD was operationalized in the U.S. A written, verified response for each S-RAD model version, including all those deployed in Beta and to the present, showing: 1) every input (“feature”) in each model; 2) a description of each input (“feature”); 3) the weight assigned to that input (“feature”), in each of its applicable models; 4) the directional impact of that input (“feature”) on an S-RAD score in that model (i.e. positive, ambiguous, negative, or any other relative terms employed by Uber); and 5) each city ID (and corresponding specific description of that city ID) where the particular beta/model and its applicable inputs (“features”) were/are deployed along with the start and end date of the deployment; A verified written explanation of how Uber calculated the average S-RAD “score” for each Bellwether trip, including whether the calculation was based on all trips or completed trips; S-RAD flagging “thresholds” and “trigger rates” set in U.S. cities from the implementation of S-RAD (in 2022) to the present; S-RAD scores (for trips on which S-RAD ran) in the produced safety data; The UUID's of all Type 1 and Type 2 “slip-through” trips in the produced safety data and produce data and corresponding information about these trips, including: a) S-RAD “nudges” that were sent as part of these trips, including tables labeled “srad_trip_nudges”, and/or b) data relating to “Ride Checks” that were sent and recorded by Uber relating to these trips, including tables labeled secure_safety.ride_check_incident_rider_log; and Provide a date for the 30(b)(6) deposition of Sunny Wong following Uber's verified answers to these questions and production of this data and related information *4 Dkt. No. 4009 at 4; see also Dkt. No. 4009-1 (elaborating on Plaintiffs' requests in a separate document, which is of questionable compliance with Pretrial Order No. 8's five-page limit for discovery disputes). Though Plaintiffs' raise a legitimate question of whether Uber should have disclosed S-RAD as a non-custodial data source earlier in litigation, they do not meaningfully dispute Uber's assertion that it has “produced thousands of S-RAD documents starting in 2024; provided written responses, per Party agreement, to several RFPs specifically about S-RAD; and produced comprehensive S-RAD data for the trips at issue.” Dkt. No. 4009 at 6. As Plaintiffs' expert Jonathan Jaffe acknowledges, some of the data that Uber has already produced related to S-RAD, includes: 33. Uber has produced the S-RAD Scores for each bellwether trip (“Bellwether Trip's S-RAD Score”). 34. Uber has produced the parameters and values Uber used to compute the Bellwether S-RAD Scores (“Bellwether Trip S-RAD Parameters”). 35. Uber has produced whether the Bellwether Trip was flagged and/or actioned by SRAD (“Bellwether Trip S-RAD Status”). 36. Uber has produced the S-RAD Trigger Rates over the 7 days preceding the Bellwether Trips only for the specific city in which the Bellwether Trip occurred (“Bellwether Trip 7-Day Preceding City Trigger Rates”). 37. Uber has produced the average S-RAD score across other trips over the 7 days preceding the Bellwether Trips only for the specific city in which the Bellwether Trip occurred (“Bellwether Trip 7-Day Preceding City Average S-RAD Score”). 38. Finally, Uber has produced the median 1-star rating of other drivers over the 7 days preceding the Bellwether Trips only for the specific city in which the Bellwether Trip occurred (“Bellwether Trip 7-Day Preceding City Driver Median 1-Star Ratings”). ECF No. 4009-2, ¶¶ 33–38. Plaintiffs have not established the same degree of evasiveness with respect to S-RAD data as discussed above with respect to Flack. The only example they provide of a purportedly misleading response that might have prevented them from pursuing this issue before the substantial completion deadline is testimony by Michael Akamine purportedly indicating that underlying data tables were “not easily accessible and may not even exist.” Dkt. No. 4009 at 5 (Plaintiffs' characterization). The exhibit Plaintiffs cite is an excerpt of Akamine's May 20, 2025 deposition in which Plaintiffs' counsel questioned him about whether Uber could determine at this point whether a particular ride was in a “control” or “treatment” group for Uber's testing of the S-RAD system. See generally Dkt. No. 4012-1 (provisionally under seal). Akamine in fact testified that he believed “the data is there somewhere,” he was “pretty sure it could be found,” but it was not something he had direct experience seeking. Id. at 14 (Akamine Dep. at 400:13–25). He went on testify that, he was “sure ... there should be some data table ... that looks at how, you know, when S-RAD is functioning and matches, that that's stored there as we take action on it.” Id. at 15 (Akamine Dep. at 401:3–5). Though Akamine stopped short of certainty that such tables are stored indefinitely, his testimony tended to indicate that the data was available, not (as Plaintiffs now assert) that it was unavailable. *5 Moreover, Plaintiffs requests for written verifications appear to address issues that could be resolved through the further deposition of Sunny Wong. Uber agreed to produce Wong for a further deposition that Plaintiffs canceled, and remains willing “to proceed expeditiously” with that deposition, id. at 8. Though Plaintiffs raise concerns about Wong's previous deposition testimony, Dkt. No. 4009 at 5 & n.6, the testimony they cite does not reflect a lack of reasonable preparation, see generally Dkt. No. 4012-2 (provisionally under seal). If Plaintiffs believed further document production was necessary after Wong's June 25, 2025 deposition, they could have raised such a request well before their August 11, 2025 email request to Uber, see Dkt. No. 4009 at 6, and presented any dispute to the Court well before the parties' September 24, 2025 joint letter, which they filed two days before a recently-extended deadline for service of expert reports, Dkt. No. 3997. The Flack data discussed above also involved requests made after the substantial completion deadline and may require supplementation of expert reports. But unlike that data, Uber has already produced extensive data regarding the S-RAD system (particularly with respect to the particular rides at issue in the bellwether cases), and Plaintiffs have not shown substantial impediments to their pursuit of this discovery before the substantial completion deadline. Plaintiffs appear to have identified S-RAD as significant since well before that deadline and have not identified new evidence of its relevance comparable to Nilles's testimony regarding the Flack data. The Court therefore orders limited relief as follows. Uber shall produce Sunny Wong for a further deposition, no later than one week from the date of this Order unless Plaintiffs agree to a later date. This Court is not available to supervise that deposition, but the Court would consider ordering the deposition reopened again if Wong's testimony is substantially evasive. Following Wong's deposition, if Plaintiffs identify specific representations by Uber that Wong (or another suitable deponent) was unable to confirm or correct through sworn testimony, Uber shall provide written verifications of those representations, or of the true facts if the representations were in any way inaccurate. Plaintiffs shall identify any such representations no later than two business days after Wong's deposition, and Uber shall provide verifications no later than three business days after Plaintiffs' identification of the representations at issue. Taking into account the likely effect of further extensive discovery at this late stage, Plaintiffs' requests are otherwise DENIED as unduly burdensome and disproportionate to the needs of the litigation. IT IS SO ORDERED.