Deposition of Vaishnavi Jayakumar (Volume 1)

Raw text

Transcribed text as recorded, including headers and formatting fragments.

Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 1 of 211 AMENDED Exhibit 2 PLAINTIFFS' OMNIBUS OPPOSITION TO DEFENDANTS' MOTIONS FOR SUMMARY JUDGMENT Case No.: 4:22-md-03047-YGR MDL No. 3047 In Re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation 1 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 UNITED STATES DISTRICT COURT FOR THE NORTHERN DISTRICT OF CALIFORNIA 2 IN RE: SOCIAL MEDIA ADOLESCENT 3 ADDICTION/PERSONAL INJURY PRODUCTS LIABILITY LITIGATION :MDL NO. :4:22-MD- :3047-YGR : Page 2 of 211 4 5 6 SUPERIOR COURT OF THE STATE OF CALIFORNIA FOR THE COUNTY OF LOS ANGELES SPRING STREET COURTHOUSE 7 COORDINATION PROCEEDING SPECIAL TITLE : (RULE 3,400) : 8 : SOCIAL MEDIA CASES : 9 :LEAD CASE :NO. FOR 10 THIS DOCUMENT RELATES TO: :FILING : PURPOSES 11 STATE OF TENNESSEE, ex rel, JONATHAN SKRMETTI, ATTORNEY GENERAL and :22STCV21355 : 12 REPORTER, : : 13 V. : : 14 META PLATFORMS, INC., and INSTAGRAM, LLC : : 15 : CASE NO. 23-1364-IV : 16 17 DEPOSITION UNDER VIDEO EXΧΑΜΙΝΑΤION OF VAISHNAVI JAYAKUMAR 18 JANUARY 30, 2025 VOLUME I 19 Videotaped deposition of 20 VAISHNAVI JAYAKUMAR, taken pursuant to notice, was 21 22 23 held at the law offices of Baker Botts, LLP, 30 Rockefeller Plaza, New York, New York, beginning at 9:24 a.m., on the above date, before Michelle L. Ridgway, a Registered Professional Reporter, Certified Court Reporter (NJ-CCR # XI02126), Certified Realtime Reporter, Certified Shorthand Reporter (CA-CSR # 14592), and Notary Public. 24 25 877-370-3377 Golkow Technologies, A Veritext Division Page 1 www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 3 of 211 Page 22 1 that? 2 A. I opted to have independent 3 counsel. 4 Ω. Okay. Do you know why they asked 5 you that question? 6 A. I think they 7 MR. SNEED: Object to form. Sorry. 8 THE WITNESS: I think they 9 suggested it might be easier for us to prepare if 10 they co-represented me, but but eventually I 11 decided to have independent counsel. 12 BY MR. WARREN: 13 Ω. I see. 14 All right. Could you please 15 introduce yourself to the jury, Ms. Jayakumar? 16 A. Sure. I'm Vaishnavi. I live in 17 18 19 20 21 22 23 New York. I currently run a product and policy advisory known as Vyanams Strategies, VYS for short. We advise companies, civil societies and governments on how to build safe equitable products for children and teens online. That ranges from social media, to gaming, to marketplace products, to AI products. I've been doing this for about a 24 year. 25 877-370-3377 Prior to this, I spent the last Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 4 of 211 Page 23 1 14 years of my career working on child safety and 2 tech policy in a variety of different roles. 3 My most recent role was the head of 4 5 6 youth policy at Meta, which I held until May of 2023. I started my time at Meta as the head of safety and well-being for Instagram specifically, 7 and then over time my portfolio evolved. 8 9 eight -- 10 11 policies at Twitter. I was the first head of Prior to Meta I had spent about eight years oh no, I spent ten years working on some of these issues. I led video 12 13 14 safety for Twitter in the Asia-Pacific region. And I worked on child safety and privacy for Google within their central team covering Asia-Pacific, 15 Middle East, Africa, and Russia. 16 And then prior to that I started my 17 career right out of college as a Imagineer at the 18 Walt Disney Company. 19 Ω. Excellent. Let's go one step even 20 before that. Where did you grow up? 21 A. I grew up in Singapore. 22 Ω. In Singapore. And where were you 23 educated? 24 A. I was educated in Singapore until 25 I until it was college time. And then I did my 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 5 of 211 Page 38 1 A. Safety first motto. I don't think 2 it is. 3 Ω. Okay. I'd like to ask you about a 4 product feature on Instagram called Reels. Are you 5 familiar with that feature? 6 A. Yes. 7 Ω. Can you explain for the jury what 8 that is? 9 A. Reels was Meta's answer to TikTok. 10 It was a series of chained short-form videos that 11 are recommended to users as they go through the 12 Reels tab. 13 Ω. Was Reels launched while you were 14 at Meta? 15 A. Yes. 16 Ω. Okay. And, again, that's an 17 Instagram feature? 18 A. It was originally on Instagram. 19 Facebook now has Reels as well. 20 Ω. Okay. In your experience, was the 21 22 23 launch of Reels on Instagram handled in a way that best protected the safety of Instagram's youngest users? 24 A. No, I don't think so. 25 Ω. And can you explain how? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 6 of 211 Page 43 1 late August/early September, did that include the 2 United States? 3 A. Yes. I believe so. 4 Ω. Okay. So let's fast forward a 5 month. 6 I'm going to hand you what we'll 7 mark as Exhibit 4. And this is a document which 8 begins with the Bates number 9 ΜΕΤΑ3047MDL-034-00329532. 10 (Document marked for identification 11 as Jayakumar Exhibit 4.) 12 BY MR. WARREN: 13 Ω. Take a minute, please, just to 14 orient yourself with that document. 15 Again, I'll direct your attention 16 to some specific parts of this. 17 But, first, is this another 18 installment of your monthly safety update? 19 A. This is another installment of the 20 biweekly update. 21 Ω. Thank you for that correction. 22 And, again, you have sent this to 23 several individuals at Meta who are identified on 24 the bcc line; is that correct? 25 A. Yes, correct. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 7 of 211 Page 44 1 Ω. Okay. Please direct your attention 2 to the top of Page 3. 3 Do you see the bullet labeled, 4 "Instagram Reels launching globally on August 5th"? 5 A. Mm-hmm. 6 Ω. Okay. Did you write that? 7 A. I did. 8 Ω. All right. Let's start near the 9 bottom of this bullet. 10 You write: 11 One continued exposure for us is 12 13 14 15 that we still do not have a sense of how minors' follow or DM graphs are changing after they go viral and the way in which this risk will be amplified when we launch globally. 16 Did I read that correctly? 17 A. Yes, you did. 18 Ω. Those were your words? 19 A. Yes. 20 Ω. Okay. Can you explain to the jury 21 what a "follow graph" is? 22 A. It refers to the kind of users that 23 are either following a minor, or that the minor is 24 25 Ω. following themselves. Okay. So maybe in layman's terms, Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 8 of 211 Page 45 1 2 it's the connections that exist for someone on Instagram? 3 A. Yeah. 4 Ω. Okay. And what does the phrase "DM 5 graphs" mean? 6 A. It's similarly the messaging 7 connections that exist for someone on Instagram. 8 Ω. Okay. And is DM short for direct 9 messaging? 10 A. Yes, direct messaging. 11 Ω. And direct messaging is a 12 communication feature built into Instagram; is that 13 correct? 14 A. Yes. 15 Ω. Okay. And can you explain to the 16 jury what your concern was as articulated in this 17 sentence? 18 A. So prior to Reels launching, I had 19 20 21 flagged that when minors, say teen users, or, you know, when young people's Reels get recommended within within the Reels platform, that's going 22 to suddenly expose them to a very wide range of 23 audiences. 24 Typically a user who posts a video 25 would have that video seen by the people they 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 9 of 211 Page 46 1 follow, if they are public accounts. Again, it 2 might be seen by the people who - sorry, the 3 people who follow them, rather. Even if it was a 4 public account, it would only be seen by the 5 accounts that already currently follow them. 6 7 8 9 10 But by introducing Reels, anyone, including people who never followed them, never knew that the account existed, would now know that the user existed, minor existed, and that they had these kind they had posted videos. 11 So that could incentivize, for 12 13 example, in the worst-case scenario, an adult predator. It could incentivize them to reach out 14 15 16 17 to the minor, try to connect with them, befriend them, try, you know, send them messages, when they previously wouldn't have even known about the minor's existence on the platform. 18 2. Okay. And you go on in this 19 section to state: 20 We had asked for this mitigation to 21 be in place following the v2 launches but have hit 22 a wall with the Data Science team working on this 23 due to limited bandwidth. 24 Did I read that correctly? 25 A. Yes, you did. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 10 of 211 Page 47 1 Ω. What was the mitigation you were 2 asking for? 3 A. So we wanted to understand whether 4 or not minors' follow or DM graphs were going to 5 change after they went viral. And by going viral, 6 7 I refer to getting a lot of views or a lot of engagement on Reels. 8 And one of the recommendations I 9 10 had made early on is that, to make this the safest experience possible, perhaps we can filter out 11 minor accounts from Reels altogether. 12 The mitigation specifically that we 13 14 had asked for was we want to measure and understand this risk. 15 16 17 I think I may have been more comfortable with us just filtering out the accounts altogether. But as a prelude to that, at least I 18 wanted to understand well, what is the level of 19 risk we are talking about. 20 And we couldn't get we couldn't 21 get that analysis done before the v2 launches. 22 Ω. And you couldn't get that analysis 23 24 25 done because Meta didn't allocate the technical resources necessary to get that analysis done, correct? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 11 of 211 Page 48 1 2 MR. SNEED: Object to form. THE WITNESS: The data science team 3 just didn't have the bandwidth to handle it. They 4 had a lot of other projects they were working on, 5 and this isn't a project they could support. 6 BY MR. WARREN: 7 Ω. Okay. And otherwise in other 8 9 words, this mitigation did not become a priority for the data science team ahead of the launch of 10 Reels? 11 12 MR. SNEED: Object to form. THE WITNESS: Yes. I would agree 13 with that. 14 BY MR. WARREN: 15 Ω. Okay. Earlier in the paragraph, 16 17 you appear to identify another safety issue with Reels, although I'd ask if you can confirm them for 18 me. But let's read the quote. 19 You say: 20 Our integrity classifiers are less 21 effective on video and not calibrated for 22 short-form video in particular (i.e., all Reels 23 content). 24 25 Did I read that correctly? A. Yes, you did. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 12 of 211 Page 54 1 have the power to fix that. 2 Ω. Okay. All you could do is flag it 3 for others which is what you did? 4 A. Yes. I think in part because 5 the -- I think I'd mentioned previously there were 6 7 a lot of limits on, you know, what I could do in my role. So the most I could do in this particular 8 context was flag it for everyone. 9 (Document marked for identification 10 as Jayakumar Exhibit 5.) 11 BY MR. WARREN: 12 Ω. Okay. I'm going to hand you what's 13 been marked as Exhibit 5. 14 This is a short one. Just let me 15 16 know when you've had a chance to review it. For the record, this is META3047MDL-020-00271442. 17 Okay. This is a chat between you 18 and from July 29, 2020; is that right? 19 A. Yes. 20 Ω. And did you send and receive these 21 messages in the ordinary course of your work at 22 Meta? 23 A. Yes. 24 Ω. Okay. And this would have been 25 right around the time of the biweekly update that 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 13 of 211 Page 55 Can you please read to the starting with the "There are a few areas around child 1 we just looked at, right? 2 A. Yes. 3 Ω. Okay. 4 5 jury your message to Ms. second sentence. 6 A. 7 8 9 10 11 12 13 endangerment, it'd be great to discuss for H2" the second half of the year. "Our existing classifiers do not work great on short-form video content, which is effectively all of Reels content. We also have identified known risks around virality that there is no current mitigation for in the Reels product roadmap namely, predatory DMs from adults sent to minors who are more easily discovered via their Reels content. +1 to suggestion to get on a call to preliminarily 's discuss this before looping in anyone specific from child safety CI." 14 15 16 17 18 19 Which refers to the central 20 integrity team. 21 Ω. Okay. So this was you, again, 22 expressing the same concern about the inefficacy of 23 the classifiers in catching harmful content on 24 Reels, correct? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 14 of 211 Page 56 1 Ω. Okay. And I see at the top of this 2 chat there's actually several other e-mail 3 addresses; is that right? 4 A. Yes. 5 Ω. So those people also would have 6 seen your chat message to Ms. ? 7 A. Yes. 8 Ω. Okay. And what was Ms. 's role 9 at the company? 10 A. I don't remember. I think she was 11 a technical program manager. 12 Ω. Okay. Ms. Jayakumar, at the time 13 14 15 Reels launched, would you agree that Meta did not have an effective way of enforcing its community standards on Reels? 16 MR. SNEED: Object to form. 17 THE WITNESS: Yes, I would agree 18 with that. 19 BY MR. WARREN: 20 Ω. Okay. Did the to the best of 21 your knowledge, did Meta warn the public that it 22 23 didn't have an effective way to implement its community standards on Reels? 24 MR. SNEED: Object to form. 25 877-370-3377 THE WITNESS: To the best of my Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 15 of 211 Page 57 1 knowledge, no. 2 BY MR. WARREN: 3 Ω. Did Meta modify its community 4 5 standards to make clear that they were not being enforced effectively on Reels? 6 MR. SNEED: Object to form and 7 foundation. 8 THE WITNESS: No, not not to the 9 best of my knowledge. 10 BY MR. WARREN: 11 Ω. Okay. Do you think someone reading 12 13 Meta's community standards in the fall of 2020 would have any way of knowing those policies were 14 not being effectively enforced on Reels? 15 MR. SNEED: Objection to form and 16 foundation. 17 THE WITNESS: I don't think so. 18 BY MR. WARREN: 19 Ω. Okay. I'm going to hand you 20 another exhibit. This will be Exhibit 7. 21 (Document marked for identification 22 as Jayakumar Exhibit 6.) 23 BY MR. WARREN: 24 Ω. And for the record this is 25 ΜΕΤΑ3047 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 16 of 211 Page 58 1 MR. WARREN: I'm sorry. This will 2 be Meta 6. Withdrawn. I'll move to strike my own 3 comment. 4 This will be Exhibit 6. 5 And for the record, this is 6 ΜΕΤΑ3047MDL-034-00037237. 7 BY MR. WARREN: 8 Ω. All right. This is a really long 9 document. You can take as much time as you need to 10 browse it, although I will be focusing you in on 11 specific areas. Just let me know when you're 12 ready. 13 A. I think we can go. 14 Ω. Okay. Do you recognize this 15 document? 16 A. Yes. I believe this is a document 17 that Antigone puts together put together for 18 Meta leadership on child safety across the 19 different products and services of Meta. 20 Ω. Okay. And the document is 21 entitled, "Child Safety State of Play (9/24)"; is 22 that right? 23 A. Yes, correct. 24 Ω. And I take it that's a reference to 25 September 24th? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 17 of 211 Page 59 1 A. Yes. Correct. 2 Ω. 3 4 Okay. I'd like your help in identifying the year of this document. And if you browse to the back, you'll see a comment history 5 that may provide some clues here. 6 A. I think this is from 2020. 7 Ω. And I'll just direct you to the 8 page ending in 37255 in case that helps. 9 A. Yes. This is from 2020. 10 Ω. Okay. And it appears that you 11 opened this conversation at some point in 2020; is 12 that right? 13 MR. SNEED: Object to form. 14 MR. WARREN: What's the basis of 15 that objection? 16 MR. SNEED: Where are you - could 17 you point me in the document where you're referring 18 to? 19 MR. WARREN: I just did. 37255. 20 MR. SNEED: 37255? 21 MR. WARREN: 55. 22 MR. SNEED: Where on the document 23 does it indicate that? 24 MR. WARREN: It says at the top, 25 "Vaishnavi J opened your conversation." 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 18 of 211 Page 60 1 MR. SNEED: The basis of the 2 objection is the mischaracterization of the 3 document given that also 4 5 6 and several others opened this conversation. MR. WARREN: How is that an 7 8 objection to my question? My question was it appears that you opened this conversation at some 9 point in 2020. 10 MR. SNEED: That's the basis of the 11 objection. 12 MR. WARREN: Okay. Are you going 13 to stand on that? 14 MR. SNEED: Sure. 15 MR. WARREN: Okay. 16 BY MR. WARREN: 17 Ω. Ms. Jayakumar, let me ask you 18 again. 19 Does it appear you opened this 20 conversation in 2020? 21 A. I - I am not sure if I did. It 22 looks like we 23 conversations in here. I don't it looks like many of us had because multiple 24 people are seen as opening the conversation. I 25 don't think so. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 19 of 211 Page 61 1 Ω. Okay. That's fine. But you've 2 seen the document before? 3 A. Yes, I have. 4 Ω. All right. So let's look at the 5 substance. I'll direct you to 37245. 6 A. Mm-hmm. 7 Ω. Do you see the heading, "Internal 8 Readiness/State of Play"? 9 A. Yes. 10 Ω. Can you read the first two 11 sentences that follow? 12 A. "Overall we are quite vulnerable 13 when it comes to age management on our apps. To 14 15 some degree this is because leadership has decided to live with certain risks at least in the near 16 term." 17 Ω. Do you agree that was the child 18 safety/state of play as of September 24, 2020? 19 A. Yes, I agree. 20 Ω. All right. And please flip to Page 21 37247, just one over. 22 Do you see the heading on this page 23 that says: 24 Not ready; work has not been 25 prioritized and more resources are needed? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 20 of 211 Page 62 1 A. Yes. 2 Ω. Okay. And a couple rows underneath 3 that heading it says: 4 Addressing virality on Instagram. 5 Do you see that? 6 A. Yes. 7 Ω. Can you read the description to the 8 jury? 9 A. "Minors who go viral on Instagram 10 11 12 13 14 will likely experience increased interactions with unknown people. As their content and by extension their account becomes surfaced to more unconnected people, they would benefit from additional protections that they can turn on, like limiting 15 16 who can send them private DMs or follow requests. However, with a lack of age data, we cannot 17 18 19 20 effectively develop solutions at scale that address virality of minors on Instagram without also generating a large number of false positives/false negatives." 21 Ω. 22 23 Okay. So this is the same issue you'd flagged in your June and July biweekly safety updates prior to the launch of Reels, correct? 24 25 A. Ω. Yes. 877-370-3377 Okay. So is it fair to say that Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 101 of 211 Page 264 1 Ω. And all the e-mail addresses 2 identified on here are official Meta e-mail 3 addresses, right? 4 MR. SNEED: Object to form. 5 THE WITNESS: Yes. 6 BY MR. WARREN: 7 Ω Who is ? 8 A. was our -- well, is our 9 Latin America safety policy representative. 10 Ω. Okay. And who is 11 ? 12 A. Our safety policy representative 13 for Sub-Saharan Africa. 14 Ω. Can you read the message Ms. 15 sent to this group, including you, at 6:45 a.m.? 16 A. "@silent is anyone else receiving 17 child abuse reports on platform? In the last 18 24 hours I have received a ton. Sad face. I've 19 20 seen terrible things, we're not detecting it. Sad face." 21 Ω. 22 Okay. And Ms. would have been received by all the individuals in 's messages 23 the to line, correct? 24 A. Correct. 25 Ω. And that includes Ms. Ravi 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 102 of 211 Page 265 1 Sinha, and ? 2 A. Yes. 3 Ω. And yourself? 4 A. Yes. 5 Ω. I'm handing you another document 6 7 that we'll mark for identification as Exhibit 30. (Document marked for identification 8 as Jayakumar Exhibit 30.) 9 BY MR. WARREN: 10 Ω. This document bears the Bates stamp 11 ΜΕΤΑ3047MDL-050-00285734. 12 Is this a chat thread among you and 13 other Meta employees from March 30, 2020? 14 A. Yes. 15 Ω. Did you participate in this chat in 16 the ordinary course of your work at Meta? 17 A. Yes. 18 Ω. Can you please tell the jury who 19 the other individuals are that are on this thread? 20 A. was the head of public 21 policy for Instagram in Europe. Karina Newton was 22 the director of public policy for Instagram. So 23 reported in to Karina. 24 25 was, I don't remember who but I think she was on the content strategy 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 103 of 211 Page 266 1 team. 2 was on the content 3 strategy team, so the team that would be able to 4 modify language that appears within the app. 5 And was Instagram 6 public policy lead for Germany. 7 8 was Instagram product policy. And I think 9 that's it. 10 Ω. 11 Okay. Now you participated in this chat in the ordinary course of your work at Meta? 12 A. Yes. 13 Ω. All right. Apologies if I already 14 asked you that. 15 This chat is two days after the 16 last document we looked at in which said 17 she received a ton of child abuse reports and seen 18 terrible things, correct? 19 A. Yes. 20 Ω. All right. The chat begins with a 21 message from Ms. ; is that right? 22 A. Yes. 23 Ω. And it appears she was flagging an 24 issue relating to the reporting of child sexual 25 abuse material on Instagram; is that right? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 104 of 211 Page 267 1 A. Yes. 2 Ω. Okay. Specifically she points out 3 that when users were reporting this content, they 4 would get a popup message; is that right? 5 A. Yes. 6 Ω. And she shares an image of that 7 message, which I'll represent to you is at the back 8 of this exhibit. Any reason to dispute that? 9 A. I don't think I have the image. 10 Ω. It should be it should have the 11 last digits 285737. Do you see that? 12 A. I don't have that. 13 MR. WARREN: All right. Let's go 14 off the record and we'll fix that. I apologize. 15 THE VIDEOGRAPHER: The time right 16 now is 3:54 p.m. We are off the record. 17 (Short break.) 18 THE VIDEOGRAPHER: The time right 19 now is 4:10 p.m. We're back on the record. 20 BY MR. WARREN: 21 Ω. Okay. Ms. Jayakumar, do you now 22 see the attachment to this exhibit? 23 A. I do. 24 Ω. Okay. And can you please read the 25 popup message to the jury? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 105 of 211 1 2 3 A. Page 268 "We couldn't review your report. We have fewer people available to review the reports because of the coronavirus (COVID-19) 4 pandemic, so we're only able to review content with 5 the most potential for harm. If you don't want to 6 see amyloves0916 on Instagram, you can unfollow, 7 8 9 mute or block them to hide their posts and comments from your feed. Reports like yours are an important part of making Instagram a safe and 10 welcoming place for everyone." 11 Ω. Okay, terrific. And just to 12 orient, Ms. was sending this popup message 13 to you and several others and flagging that this 14 15 was showing up in response to reports of child sexual abuse material on Instagram; is that right? 16 A. Yes. That's right. 17 Ω. Okay. And so then you and several 18 others discuss what to do about this. 19 Karina Newton at 9:11 a.m. says: 20 Weird that we sent this back for 21 CEI. 22 Did I read that correctly? 23 A. Yes, correct. 24 Ω. Okay. And is CEI another term for 25 child sexual abuse material? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 106 of 211 Page 269 1 A. Yes. 2 Ω. Okay. And does it stand for child 3 exploitation imagery? 4 A. Yes. 5 Ω. Okay. Did you also think it was 6 7 weird that Instagram would be telling people that it couldn't review reports of child sexual abuse 8 material? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes, that was odd. 11 BY MR. WARREN: 12 Ω. Okay. And then at 9:29 a.m., 13 writes: 14 I think the issue here is that we 15 shouldn't be auto-closing CEI. 16 Is that right? 17 A. Yes, that's right. 18 2. So that's a somewhat different 19 concern, right? 20 A. Yes. 21 Ω. Ms. isn't expressing a concern 22 23 with what Instagram tells users who report child sexual abuse material. She's expressing a concern 24 with what Instagram actually does in response to 25 those reports. Is that fair? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 107 of 211 Page 270 1 MR. SNEED: Object to form and 2 foundation. 3 THE WITNESS: Yes, correct. 4 MR. WARREN: And, Mr. Sneed, I'd 5 6 ask that you let me finish my question before you lodge your objection, please. 7 MR. SNEED: Do my best. 8 BY MR. WARREN: 9 Ω. Couple minutes later you chime in, 10 11 you ask whether this could be a function of how the reporter reported it. Maybe they reported it as 12 something other than CEI. 13 Do you see that? 14 A. Yes. 15 Ω. And what did you mean? 16 A. Well, it could have been that, you 17 know, it was actually CEI or CSAM and they reported 18 it as you know, maybe they reported it as 19 bullying and harassment or hate or graphic content. 20 Ω. Okay. And, again, CSAM is 21 shorthand for child sexual abuse material, right? 22 A. Yes, correct. 23 Ω. Okay. But Ms. Newton responds to 24 you and seems to say that's actually not the source 25 of the problem. 877-370-3377 Is that fair? Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 108 of 211 1 2 Page 271 MR. SNEED: Object to form. THE WITNESS: That's what she says, 3 yes. 4 BY MR. WARREN: 5 Ω. And she says, specifically: 6 It doesn't actually matter what 7 people chose in the reporting flow. We run 8 classifiers to categorize. 9 Do you see that? 10 A. Yes. 11 Ω. And you say, "Ahh, I see," correct? 12 A. Yes. 13 Ω. Okay. And how did you understand 14 Ms. Newton's clarification there? 15 A. I understood it to mean that our 16 classifiers are proactively classifying this 17 content as CEI and then would remove it. So the 18 reason why someone reports it really shouldn't 19 matter. 20 Ω. Okay. And since it's been a few 21 hours, can you remind the jury again what 22 classifiers are? 23 A. Classifiers are machine learning 24 algorithms that proactively look for - in this 25 case, integrity classifiers are machine learning 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 109 of 211 1 2 Page 272 algorithms that actively look for content that could be violating any one of our policies. 3 Ω. Okay. So am I correct that 4 5 6 7 Ms. Newton was essentially saying it doesn't matter how people report this, Instagram is using our own technology to figure out what it is and whether it's child sexual abuse material, is that fair? 8 MR. SNEED: Object to form. 9 THE WITNESS: Yes. Correct. 10 BY MR. WARREN: 11 Ω. And so please flip to the next 12 page. 13 Do you see your message at 14 9:40 a.m.? 15 A. Yes. Correct. 16 Ω. Can you please read that to the 17 jury: 18 A. "If CEI is determined by 19 classifiers, not the queue it's reported in, then 20 21 22 23 it seems we'd always have this issue. During this time where we're only reviewing the most severe harms, have we considered specific reporting flows for those harms?" 24 Ω. And what did you mean? 25 A. So Karina was saying, and I think Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 110 of 211 Page 273 I think let me 1 earlier on, I think 2 see. 3 4 5 So Karina was saying that regardless of the queue you report it in, our classifiers are automatically detecting and 6 labeling content as CEI. 7 And so then I said, if that's the 8 9 10 11 12 13 14 15 case, then we would always be Where, you know, it sounds like what might be happening is the classifier is not marking those pieces of content as CEI and then is sending because it's not CEI, isn't, you know, sending them for review or removing them. Instead it's sending the reporting person a message saying sorry, we couldn't review your report. get this issue. - and 16 Ω. So just so we have a clear record, 17 18 19 20 Ms. Jayakumar. Were you expressing a concern that Instagram's classifiers were not accurate enough to remove child sexual abuse material on the Instagram platform? 21 MR. SNEED: Object to form. 22 23 that this issue 24 I was saying I was saying that this issue wouldn't go away if the classifier was at fault. THE WITNESS: I was - 25 If this was an issue of the classifier being at 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 111 of 211 Page 274 1 fault, then this was this issue wasn't going to 2 3 4 go away. And so I was suggesting that we offer a specific reporting option instead. So a specific reporting option for CEI. 5 BY MR. WARREN: 6 Ω. Okay. And so as of this moment in 7 8 9 time, March 30, 2020, is it correct that Instagram did not have a specific way for people to report child sexual abuse material on the platform? 10 MR. SNEED: Object to form. 11 THE WITNESS: Yes. 12 BY MR. WARREN: 13 Ω. Okay. And that would have been 14 true whether the user was actually a user of 15 Instagram let me withdraw the question. 16 And that would have been true 17 whether the person doing the reporting was logged 18 into Instagram or not; is that correct? 19 MR. SNEED: Object to form. 20 THE WITNESS: Yes. That's right. 21 BY MR. WARREN: 22 Ω. Okay. And can you read to the jury 23 Ms. 's response to your question? 24 A. "We don't have an in-app reporting 25 option for CEI that I'm aware of." 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 112 of 211 Page 275 1 Ω. And can you read to the jury 2 Ms. Newton's response to your question? 3 A. "We do not." 4 Ω. Okay. And can you read to the jury 5 Ms. 's further response at 9:43 a.m.? 6 A. "I don't think updating our 7 8 9 10 reporting flow options is minimal product work. And I'd be concerned that people would abuse this reporting option to report anything they want reviewed and we'd need to review it because it's 11 been reported as CEI." 12 Ω. So did you interpret Ms. 13 14 15 16 to essentially be saying, A, it'd be a ton of work to build a reporting option for child sexual abuse material, and B, if we did that, then anything that came through that channel, we'd have to review, 17 which would be even more work? 18 MR. SNEED: Object to form. 19 Characterization. 20 THE WITNESS: Yes. That's how I 21 22 23 24 25 interpreted her message. I was kind of remember this conversation now, and it was very surprising to me. Because reporting CEI is sort of baked into most platforms. It was a surprise to me, I was only two months into the company, it was - I 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 113 of 211 Page 276 1 a surprise to me that you couldn't report CEI 2 within the app. 3 BY MR. WARREN: 4 Ω. Did you feel like Instagram should 5 have developed a reporting option for reporting 6 child sexual abuse material? 7 A. 8 9 Yes, I made this recommendation multiple times. I think, you know, when I joined I made this recommendation here in this chat. I made 10 11 it several times in conversations after that. Eventually I believe Instagram did report 12 13 14 15 include the option. But there was a lot of pushback along these lines, that it would be significant product work and that people might abuse the reporting option. 16 Ω. Does Ms. 's response strike 17 you as putting the safety of kids first? 18 MR. SNEED: Object to form. 19 THE WITNESS: It does not. 20 BY MR. WARREN: 21 Ω. Okay. And during this time, 22 Instagram had options for reporting other kinds of 23 content that violated its policies, like spam, 24 right? 25 A. Yes. Exactly. There were many 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 114 of 211 Page 277 1 other issues that could be reported in-app, but not 2 CEI. 3 Ω. Okay. So you have no doubt that 4 5 6 Instagram had the technological ability to create an option for people to specifically report child sexual abuse material? 7 MR. SNEED: Object to form. 8 THE WITNESS: Yes. It is not a 9 10 significant amount of product work. It's you essentially add an additional option to the 11 12 existing number of reporting, you know, options that are out there. 13 It is significant work to integrate 14 15 16 17 that into the reviewing queues and then take a look at the reports that have come in. But the actual work of introducing a reporting option is not significant. 18 BY MR. WARREN: 19 Ω. Okay. Let me direct your attention 20 to the message you sent at 9:50 a.m. 21 A. Okay. 22 Ω. Could you read that to the jury, 23 please? 24 A. "If reports are being categorized 25 using classifiers, I'm not sure that re-reporting 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 115 of 211 Page 278 1 2 3 4 the content is going to change that. The best option is one where a classifier just doesn't get it wrong, but where it does, re-reporting may not help." 5 Ω. Okay. And were you responding here 6 7 8 9 to the suggestion made earlier in the chat that maybe Instagram should just make people report child sexual abuse material again after receiving that popup? 10 11 MR. SNEED: Object to form. THE WITNESS: Yeah, the earlier 12 13 14 15 16 17 conversation was essentially suggesting that instead of saying we're not able to review your report, to actually urge people to report the content again, if they feel like it is a serious harm like CEI. And I was saying that, well, if the 18 classifier is not detecting it as CEI now, 19 20 re-reporting it is not going to suddenly make the classifier detect that as CEΙ. 21 BY MR. WARREN: 22 Ω. Okay. So if that suggestion had 23 gone into effect, people only would have had to 24 report spam once, but they would have had to report 25 child sexual abuse material twice in order for 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 116 of 211 Page 280 1 material was zero tolerance, correct? 2 A. 3 Ω. Okay. 4 5 Yes. That was a problem. So if the classifier wasn't actually working, there'd be a disconnect between what was on the platform and what the policy was, 6 correct? 7 MR. SNEED: Object to form. 8 Characterization. 9 THE WITNESS: Yes, that's right. 10 BY MR. WARREN: 11 Ω. Okay. I'm going to hand you 12 another exhibit. And this will be marked as 13 Exhibit 31, and it's a document bearing the Bates 14 stamp META3047MDL-144-00000324. 15 (Document marked for identification 16 as Jayakumar Exhibit 31.) 17 BY MR. WARREN: 18 Ω. Okay. Please turn to the third 19 page of this. 20 Do you see your name in the middle 21 of the page? 22 A. Yes. 23 Ω. And does that indicate to you that, 24 while you were working at Meta, you saw this 25 conversation? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 117 of 211 Page 281 1 A. Yes. I commented on the 2 conversation. 3 Ω. All right. Very well. 4 And this appears to be a 5 September - I'm sorry, I apologize, withdrawn. 6 This document appears to be dated 7 June 16, 2020. Do you agree with that? 8 A. Yes. 9 Ω. And the title of the document is, 10 "Instagram Test User: Minor Sexualization/CEI 11 Discoverability." 12 Do you agree with that? 13 A. Yes. 14 Ω. Can you explain what this document 15 is? 16 A. So was on the 17 18 19 20 21 22 23 24 in some way 25 investigations team, if I'm remembering the name correctly, but had essentially run a test to see the likelihood of being surfaced minor content sexualizing minors in certain environments. So she created an Instagram test user, followed about 70 accounts that I think were either sexualizing minors or talking about minors actually it doesn't specify which accounts she followed. So I don't know. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 118 of 211 Page 282 But the result of the test was 1 2 that, had once she followed those accounts, the 3 Feed, Explore, and Recommended accounts were filled 4 almost exclusively with minor sexualization content 5 and accounts. 6 Ω. Okay. And it appears there were 7 some other results as well, right? 8 A. Yes. The the user - the test 9 account that they set up also received over 20 10 follow requests. 11 And in the top results of the 12 13 14 Explore section, the Explore tab which is a recommendation surface, they found CEI being recommended. 15 Ω. Okay. So just to summarize. 16 Investigators at Instagram set up a test account, 17 and in less than 24 hours of setting up that 18 account, the Feed, Explore, and Recommended 20 19 Accounts sections were filled with minor sexualization content, and child sexual abuse 21 material was recommended to the account as the top 22 result in Explore, correct? 23 A. Correct. 24 MR. SNEED: Object to form. 25 Characterization. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 119 of 211 1 2 3 4 5 a 6 7 Page 286 minors behaving in a very sexualized way. It could be just, you know, a skimpy outfit. It could be a provocative dance. But it wouldn't necessarily meet the threshold of being CEI. So it's sort of -- if you want to think about it, the closest analogy is minor sexualization is the borderline policy for CEI. 8 BY MR. WARREN: 9 Ω. Can you go back to the other 10 exhibit, which is Exhibit 31. 11 A. Yes. 12 Ω. And can you read that third bullet 13 under the TL;DR section. 14 A. "CEI being recommended as the top 15 result in the Explore section." 16 Ω. Okay. So the investigation 17 indicated that child sexual abuse material was 18 being recommended in Explore, correct? 19 A. Yes. 20 Ω. And that is inconsistent with the 21 22 zero tolerance policy that had been in place for Instagram, at least since April 2020, correct? 23 MR. SNEED: Object to form. 24 Characterization. 25 877-370-3377 THE WITNESS: Yes. Correct. Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 120 of 211 Page 287 1 BY MR. WARREN: 2 Ω. 3 side. You can put that document to one I'm going to hand you the next exhibit which 4 is Exhibit 32. 5 (Document marked for identification 6 as Jayakumar Exhibit 32.) 7 8 MR. WARREN: And this is a document bearing the Bates stamp META3047MDL-047-01219676. 9 BY MR. WARREN: 10 Ω. Have you had a chance to review 11 Exhibit 32? 12 A. Yes. 13 Ω. This appears to be a September 28, 14 2020, chat thread among you, and 15 do you agree? 16 A. Yes. 17 Ω. Who is Ms. ? 18 A. was a data scientist on the 19 child safety team. 20 Ω. And who is Mr. ? 21 A. Also on the child safety team as 22 well. 23 Ω. Okay. Did you participate in this 24 chat in the ordinary course of your work at Meta? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 121 of 211 Page 288 1 Ω. Okay. So September 28, 2020, is 2 three months roughly after the last document we 3 looked at, correct? 4 A. Yes. 5 Ω. Okay. And it's several months 6 7 8 9 after the document in which you had raised the possibility that Instagram's classifier for child sexual abuse material might have some problems; is that right? 10 MR. SNEED: Object to form. 11 THE WITNESS: Yes. Yes. 12 BY MR. WARREN: 13 Ω. And at 1:02 p.m., Ms. writes 14 to the group and says: 15 Our CEI deletes on Instagram versus 16 Facebook are approximately 21 times fewer. 17 Did I read that correctly? 18 A. Yes. 19 Ω. What does that statistic mean? 20 A. It means that, if you look at the 21 amount of CEI being removed proactively on 22 Instagram, versus Facebook, Instagram has far fewer 23 number of CEI deletions. 24 Ω. Okay. And then Ms. goes on 25 to provide her own explanation. At 1:04 p.m. she Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 122 of 211 Page 289 1 says: 2 Basically our enforcement overall 3 on Instagram is not as good and prone to errors we 4 have less experience with. 5 Did I read that correctly? 6 A. Yes. 7 Ω. Okay. Did you agree with that 8 statement? 9 A. Yes. 10 Ω. And then Ms. continues with: 11 Another stat: We are autoenforcing 12 90 times fewer IIC on Instagram than on Facebook. 13 Did I read that correctly? 14 A. Yes. 15 Ω. Does IIC stand for inappropriate 16 interactions with children? 17 A. Yes. 18 2. 19 A. 20 removing or So what does that statistic mean? That means that we're automatically automatically removing 90 times less 21 amount of inappropriate interactions with children 22 on Instagram than on Facebook. 23 Ω. And that's not because there was 90 24 times less of that content on Instagram versus 25 Facebook, correct? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 123 of 211 Page 296 1 you just testified? 2 MR. SNEED: Object to form. 3 THE WITNESS: Yes. That's right. 4 BY MR. WARREN: 5 Ω. Okay. I'm going to hand you the 6 next exhibit. This will be Exhibit 33. It's a 7 document bearing the Bates number 8 ΜΕΤΑ3047MDL-031-00192305. 9 (Document marked for identification 10 as Jayakumar Exhibit 33.) 11 BY MR. WARREN: 12 Ω. This appears to be an internal 13 presentation prepared by Meta employees titled, 14 "Child Safety on Instagram, Actor behaviors and 15 opportunities?" 16 Do you agree with that? 17 A. Yes. 18 Ω. And I'll represent to you that this 19 was produced to us by Meta. Any reason to dispute 20 that? 21 A. No. 22 Ω. All right. Before we talk about 23 the presentation, can you turn to the page ending 24 in 2337? 25 A. Okay. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 124 of 211 Page 297 1 Ω. All right. Do you see your name 2 3 listed at the bottom of the page as a point of contact? 4 A. Yes. 5 Ω. Okay. Do you recall seeing this 6 presentation before? 7 A. Yes. 8 Ω. Okay. And at the back of this 9 document is a 10 sheet containing its metadata, and I'll represent to you the metadata we were provided 11 by Meta's counsel indicates the document was 12 13 created on October 2, 2020. Do you have any reason to dispute that? 14 A. No. 15 Ω. Okay. So let's go to the first 16 page after the title page. 17 Can you read to the jury the goal 18 of this presentation? 19 MR. SNEED: I'm just going to 20 object to the characterization of the created date. 21 But I understand what you're referring to. 22 Go ahead. 23 THE WITNESS: "To summarize and 24 25 present early data explorations on bad actors who seek to interact inappropriately with or harm 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 125 of 211 Page 298 1 minors on Instagram." 2 BY MR. WARREN: 3 Ω. Okay. And there's a trigger 4 5 A. warning at the bottom. Can you read that? "This deck includes images that 6 7 illustrate bad actors behaviors. These include (but are not limited to) IIC, CEI, and MS." 8 Ω. IIC is inappropriate interactions 9 with children? 10 A. Yes. 11 Ω. CEI is child exploitation imagery? 12 A. Yes. 13 Ω. And MS is minor sexualization, 14 right? 15 A. Yes. 16 Ω. Is it fair to say the authors of 17 18 this presentation thought a warning was necessary, given that sort of content was contained in the 19 document? 20 A. Yes. 21 MR. SNEED: Object to form and 22 foundation. 23 BY MR. WARREN: 24 Ω. And do you agree that's a 25 responsible thing to do because, generally 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 126 of 211 Page 299 1 speaking, people don't want to be exposed to sexual 2 content involving kids, right? 3 A. Yes. 4 Ω. All right. Can you flip to the 5 slide ending in 2308. 6 Is this a summary slide? 7 A. Yes. 8 Ω. Can you read to the jury Point 5 on 9 the summary slide? 10 A. "We are only at the beginning of 11 our journey. We have a long road ahead of 12 13 understanding, sizing, detecting, filling gaps in policy, developing solutions, and ensuring 14 consistent and appropriate enforcement." 15 Ω. 16 17 18 19 Okay. Would you agree this is consistent with the e-mail thread we just looked at, in which you indicated there were detection, measurement, and enforcement challenges related to child sexual abuse material? 20 A. Yes. 21 Ω. Please turn to the next slide. 22 Can you read the headline? 23 A. "Data on the size of the problem 24 and gaps." 25 Ω. 877-370-3377 Can you read the data underneath? Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 127 of 211 Page 300 1 A. 2 compared to Facebook. 38 times more victims on Instagram Yet, 21 times fewer CEI 3 deletes compared to Facebook. And 90 times fewer 4 auto-enforcing of inappropriate interactions with 5 children, compared to Facebook. 6 Ω. Now, there's a ton of info in this 7 presentation. But I'd like to turn to some 8 examples that are raised in the presentation. 9 And that may be hard, which is why 10 the trigger warning is on the presentation. So I 11 apologize to you that we have to do it. 12 Can you please turn to the page 13 ending in 2324. 14 15 A. What's the title of the slide? "Accounts Sexualizing Teens." 16 Ω. And do you agree that what's 17 displayed underneath are, in fact, images on 18 Instagram that sexualize teens? 19 MR. SNEED: Object to form. 20 THE WITNESS: Yes. 21 BY MR. WARREN: 22 Ω. Including one account titled 23 "Beautiful Little Models"? 24 A. Yes. 25 Ω. And one account titled "Teen Bikini Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 128 of 211 Page 301 1 Pic"? 2 A. Yes. Yes. 3 Ω. Let's turn to the next slide. It's 4 called, "Searching and Discovering," right? 5 A. Yes. 6 Ω. Can you read the subtitle? 7 A. "As part of Project Gumshoe, 8 community operations and" I forget what OS is -- 9 10 "CO and OS investigators collaborated with Central Integrity to create a test account that simulated 11 12 the bad actor discoverability process. Below are some recommendations and suggestions for this 13 account." 14 Ω. And when it says recommendations 15 16 and suggestions, it's referring to what the Instagram algorithm recommended and suggested to 17 the test account, correct? 18 A. Yes. 19 Ω. And can you describe what was 20 recommended and suggested to the account? 21 MR. SNEED: Object to form. 22 THE WITNESS: Accounts that 23 referenced young boys and young children, as well 24 as megalinks, which is a form of file transfer that 25 takes place off platform and has been used in many 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 129 of 211 Page 302 instances to transfer CEI between individuals. 1 2 BY MR. WARREN: 3 Ω. 4 In the top left box, it appears there's an image of someone just typing the letter 5 H into a search bar. Do you agree with that? 6 A. Yes. 7 Ω. What's the first account that 8 Instagram recommends? 9 MR. SNEED: Object to form. 10 THE WITNESS: "Hottieskids." 11 BY MR. WARREN: 12 Ω. And what is the description of the 13 account? 14 A. "Hot young boys." 15 Ω. And let's look at the picture to 16 the right, which is called "Discover People." 17 What's that feature on Instagram? 18 A. It's accounts that you may want to 19 follow or that you may find interesting based on 20 your interests and people you follow. 21 Ω. And Instagram affirmatively 22 suggests accounts to users through that feature, 23 right? 24 A. Yes. 25 Ω. And what did it suggest to this Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 130 of 211 Page 303 1 account? 2 MR. SNEED: Object to form. 3 4 THE WITNESS: It seemed to recommend a number of accounts related to young 5 children. 6 BY MR. WARREN: 7 Ω. Including an account called 8 "hotcutes"? 9 A. Yes. 10 Ω. And in this screen capture, one of 11 the accounts appears to end in "boy745," and 12 13 include a picture of a young boy with no shirt, right? 14 A. Boy745 oh yes, that's right. 15 Ω. Can you describe the screenshot to 16 the right of that? 17 A. This looks like the Explore tab of 18 19 this account, and the videos and images being recommended are all of young boys, many of whom are 20 shirtless or presented in a sexualizing way. 21 Ω. Let's flip to the next page. 22 What's the title of the slide? 23 A. "Soliciting Images." 24 Ω. Can you read the subheading please: 25 A. "Lower risk if soliciting from 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 131 of 211 Page 304 1 other actors. 2 3 4 Higher risk if soliciting from targets (see below slides on Active/Higher effort behaviors). This behavior is illegal and reportable to NCMЕС. It's common for actors to 5 pose as teens." 6 Ω. And is it your understanding of 7 what follows, that these are examples of adults 8 posing as teens to sexually exploit kids on 9 Instagram? 10 11 MR. SNEED: Object to form. Characterization. 12 THE WITNESS: It's not clear if 13 these are adults posing as teens. But it does look 14 like accounts claiming to be teens. 15 BY MR. WARREN: 16 Ω. Accounts claiming to be teens and 17 soliciting -- 18 A. Soliciting sexual images, yes. 19 Ω. Okay. And that includes an account 20 that says, "DM me I am 15"? 21 A. Yes. 22 Ω. And underneath, the hashtags 23 include, gay 12 years old. And that's misspelled, 24 correct? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 132 of 211 Page 305 1 Ω. And is it your understanding that 2 hashtags are commonly misspelled to avoid 3 detection? 4 A. Yes. They commonly are misspelled 5 to avoid detection, but there needs to be some sort 6 of consensus within the community to actually make 7 them valuable hashtags. 8 Ω. 9 And underneath it says, "DM me I'm 14 bottom looking for daddy or BBC, correct? " 10 A. Yes. 11 Ω. Does that have any place on 12 Instagram? 13 A. No, not at all. 14 Ω. But it was on Instagram, right? 15 A. Yes. 16 Ω. Let's turn to the next page. 17 What's the title of this slide? 18 A. "Hashtags." 19 Ω. And what's the subheading? 20 A. "Using Hashtags to find (and also 21 share) content." 22 Ω. And underneath there's a screen 23 shot with a search bar. Do you agree? 24 A. Yes. 25 Ω. And it appears the investigator has Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 133 of 211 Page 306 1 typed "childporn" into the search bar. Do you 2 agree? 3 A. Yes. 4 Ω. And what are the hashtag results 5 that Instagram recommends in response to that 6 search query? 7 MR. SNEED: Object to form. 8 THE WITNESS: Plenty of hashtags 9 with the term "child porn" in it. 10 BY MR. WARREN: 11 Ω. Including one that says, "child 12 porn is good," right? 13 A. Yes. 14 Ω. And one that says, "child porn 15 collection"? 16 A. Yes. 17 Ω. Let's skip to the slide ending in 18 2332 please. 19 What's the title of this slide? 20 A. "Comments." 21 Ω. And how would you describe the 22 content that's depicted on this slide? 23 A. They are all comments on images of 24 young girls and they are some of them are just 25 very appreciative, but many of them are fully 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 134 of 211 Page 307 1 sexualizing. 2 Ω. 3 Including one comment that says, "I'm going to slide right in, right? " 4 A. Yes. 5 Ω. And the screen shot in the middle 6 appears to display the account of username 7 trade_mega_link1. 8 Do you see that? 9 A. Yes. 10 Ω. The description says, "Trade and 11 12 exchange boy or girl mega," and the word "trade" is misspelled. 13 Do you see that? 14 A. Wait. Where am I looking? Is it 15 on this page? 16 Ω. Mm-hmm. 17 A. I don't see that. 18 Ω. All right. Based on your 19 experience of working on digital child safety 20 issues, is "mega" a common misspelling of image 21 used by pedophiles? 22 A. 23 It can be. It's usually more an abbreviation for megalinks, a platform where CEI 24 images and links are shared. 25 Ω. 877-370-3377 All right. Let's go to the slide Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 135 of 211 Page 308 1 ending in 2334. Do you see that? 2 A. Yes. 3 Ω. What's the title of this slide? 4 A. "Sharing content & connecting." 5 Ω. 6 And here, do you see a screen shot displaying the account of username 7 trade_mega_link1? 8 A. Yes. 9 Ω. All right. And the description 10 here says, "trade and exchange boy or girl mega, " 11 correct? 12 A. Megalink, yes. 13 Ω. Okay. And the account appears to 14 have several photos of young shirtless boys, right? 15 A. Yes. 16 Ω. Including one boy who appears to be 17 sleeping? 18 A. Yes. 19 Ω. And on the right, are those 20 21 examples of pedophiles messaging one another on Instagram, that the authors of this presentation 22 included as an example? 23 A. Yes. These are child sexual abuse 24 25 predators messaging one another, essentially trying to build community and trade images. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 136 of 211 Page 309 1 Ω. Let's turn to the next page. And 2 this will be the last one from this document. 3 The slide is called "User 4 Experience," right? 5 A. Yes. 6 Ω. Would you agree this appears to 7 relay the impressions provided to Instagram by 8 three of its users? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes. 11 BY MR. WARREN: 12 Ω. 13 A. Can you read them to the jury? "Seeking. 'We had an account 14 15 16 17 called "High school girls robotics team" and we started to get some older men following us and watching us. I realized that they probably just searched 'high school' or 'girls' to find us." 18 From a teen woman. 19 "Commenting. 'She posted a picture 20 21 in a bikini and there were all these older men making inappropriate comments and asking when she 22 would post more pictures. Mom speaking on 23 daughter's/influencer's experience." 24 "Stalking. 'I'll see a random 25 person watching my Story and when I look into them 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 137 of 211 Page 310 1 2 3 I realize it's an old man with kids and a wife. What business does he have watching a young girl's Story? Teen woman." 4 Ω. Okay. And actually let's do two 5 pages forward. 6 And then we'll be done with this document. Do you see the slide called 7 "Contacting"? 8 A. Yes. 9 Ω. And do you see the grey box at the 10 bottom? 11 A. Yes. 12 Ω. Can you please read that to the 13 jury? 14 A. "Disrupt the ability for predators 15 to find targets on Instagram: Facebook disrupts 16 17 18 the ability for predators to discover minors (using People You May Know) to prevent unconnected adults from discovering or initiating conversations with 19 minors. Instagram has no similar logic for 20 21 Accounts You May Follow. We're in the process of understanding minor discoverability on selected 22 23 surfaces on Instagram (example: search, suggested accounts) For more on proactive detection and 24 product solutions, refer to: 'Child safety 25 landscape Instagram' (POC, point of contact, 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 138 of 211 Page 311 1 Vaishnavi J." 2 Ω. 3 A. 4 Ω. And that's you, right? That's me, yes. Okay. What are the Accounts You 5 May Follow feature? 6 A. Accounts You May Follow so on 7 8 9 10 11 12 13 14 15 Facebook, you know, it's really centered on finding your friends. On Instagram it's not necessarily the case. You may follow a celebrity chef, you may follow a gardener. And so the concept of recommending people you may know on Facebook, on Instagram translates into accounts that you may want to follow or that you may be interested in. When I joined Instagram, I think it 16 17 18 19 20 was one of the first things that I flagged for the child safety team saying that hey, Accounts You May Follow is surfacing minors on Instagram. So it was one of the first things that I noticed, you know, as part of my assessment of the platform. 21 And so, this kicked off this 22 project to really understand where are we able to 23 easily detect minors on the platform. And then 24 25 this document, "Child safety landscape captured some of those surfaces. Instagram" Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 139 of 211 Page 312 Okay. Earlier we looked at a slide 1 Ω. 2 that had an image labeled "Discover People." 3 Do you recall that? 4 A. Yes. 5 Ω. Is that the same as Accounts You 6 May Follow? 7 A. Yes. 8 Ω. Okay. And through Accounts You May 9 Follow, Instagram affirmatively recommends certain 10 accounts to its users, correct? 11 A. Yes. 12 Ω. And you're saying that shortly 13 after you joined, you flagged the problem that 14 Accounts You May Follow may be surfacing pedophile 15 accounts to users, right? 16 MR. SNEED: Object to form. 17 THE WITNESS: I I flagged that 18 19 accounts you may know may be surfacing minor accounts to users. So teen accounts to users. 20 BY MR. WARREN: 21 Ω. I see. And this presentation was 22 from October 2020, that's ten months after you 23 joined, right? 24 A. Yes. 25 Ω. 877-370-3377 And as of ten months after you Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 140 of 211 1 2 3 Page 313 joined, Instagram had not fixed the problem that you had identified of Accounts You May Follow surfacing minor accounts to adults, correct? 4 5 MR. SNEED: Object to form. THE WITNESS: Correct, yeah. So 6 Facebook 7 as it says here, Facebook already had disrupted this ability. So minors weren't 8 surfacing in account - in People You May Know, SO 9 that potential predators wouldn't be able to find 10 them. Instagram had not yet done that. 11 BY MR. WARREN: 12 Ω. And we looked at some of the 13 results of that earlier in this presentation, 14 correct? 15 A. Yes. 16 MR. SNEED: Object to form. 17 BY MR. WARREN: 18 Ω. Do you agree that the things we 19 just looked at are hard to stomach? 20 A. Yes. 21 Ω. Is any of it consistent with Meta's 22 public-facing policy of zero tolerance for child 23 exploitation? 24 MR. SNEED: Object to form. 25 877-370-3377 THE WITNESS: No. It's not. Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 141 of 211 Page 314 1 BY MR. WARREN: 2 Ω. 3 Did Meta at any point caveat its public-facing policy by identifying these problems 4 to the public? 5 MR. SNEED: Object to form and 6 foundation. 7 THE WITNESS: I think Meta has 8 9 10 talked on occasion and regularly about the challenges with finding some of this content. It's, you know, engaged a number of partnerships 11 with industry. 12 But I don't think it's ever really articulated the extent of the the severity of 13 the problem on the platform. 14 BY MR. WARREN: 15 Ω. And we looked at the policy, which 16 was consistent from 2020 to 2025, correct? 17 A. Yes. 18 Ω. And at no point did Meta edit that 19 policy to say, we have a zero tolerance policy that 20 we have a problem enforcing, correct? 21 MR. SNEED: Object to form. 22 THE WITNESS: Yes. 23 BY MR. WARREN: 24 Ω. Did Meta warn parents of these 25 dangers to their children on Instagram? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 142 of 211 Page 315 1 MR. SNEED: Object to form. 2 THE WITNESS: Yes. 3 BY MR. WARREN: 4 Ω. You believe they warned 5 A. Oh no, sorry, they didn't warn 6 parents. 7 Ω. Let me ask the question again to 8 get a clean record. 9 Did Meta warn parents of these 10 dangers to their children on Instagram? 11 MR. SNEED: Object to form. 12 THE WITNESS: No. 13 BY MR. WARREN: 14 Ω. Did Meta warn children they may be 15 exposed to these sorts of interactions? 16 MR. SNEED: Object to form. 17 THE WITNESS: No. 18 BY MR. WARREN: 19 Ω. But Meta thought it was worth 20 including a trigger warning on this presentation 21 for its own employees, correct? 22 MR. SNEED: Object to form. 23 THE WITNESS: Yes. 24 BY MR. WARREN: 25 Ω. All right. Let's look at another Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com 1 2 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 CERTIFICATE Page 143 of 211 Page 326 3 4 5 6 I HEREBY CERTIFY that the witness was duly sworn by me and that the deposition is a true record of the testimony given by the witness. It was requested before completion 7 of the deposition that the witness, VAISHNAVI JAYAKUMAR, have the opportunity to read and sign the deposition transcript. 8 9 Michelle L. Gray 10 MICHELLE L. RIDGWAY, 11 Certified Shorthand Reporter, (CA-CSR # 14592) 12 (Certified Court Reporter (NJ-CCR # XI02126) 13 Registered Professional Reporter, Certified Realtime Reporter 14 and Notary Public Dated: February 4, 2025 15 16 17 18 19 20 21 22 23 24 25 (The foregoing certification of this transcript does not apply to any reproduction of the same by any means, unless under the direct control and/or supervision of the certifying reporter.) 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com 1 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 144 of 211 Page 327 INSTRUCTIONS TO WITNESS 2 3 4 5 6 7 Please read your deposition over carefully and make any necessary corrections. You should state the reason in the appropriate space on the errata sheet for any corrections that are made. After doing so, please sign the 8 errata sheet and date it. 9 10 You are signing same subject to the changes you have noted on the errata sheet, which 11 will be attached to your deposition. 12 It is imperative that you return 13 14 the original errata sheet to the deposing attorney within thirty (30) days of receipt of the 15 16 deposition transcript by you. If you fail to do SO, the deposition transcript may be deemed to be 17 accurate and may be used in court. 18 19 20 21 22 23 24 25 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 145 of 211 Exhibit 2B PLAINTIFFS' OMNIBUS OPPOSITION TO DEFENDANTS' MOTIONS FOR SUMMARY JUDGMENT Case No.: 4:22-md-03047-YGR MDL No. 3047 In Re: Social Media Adolescent Addiction/Personal Injury Products Liability Litigation 1 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 UNITED STATES DISTRICT COURT FOR THE NORTHERN DISTRICT OF CALIFORNIA 2 IN RE: SOCIAL MEDIA ADOLESCENT 3 ADDICTION/PERSONAL INJURY PRODUCTS LIABILITY LITIGATION :MDL NO. :4:22-MD- :3047-YGR : Page 146 of 211 4 5 6 SUPERIOR COURT OF THE STATE OF CALIFORNIA FOR THE COUNTY OF LOS ANGELES SPRING STREET COURTHOUSE 7 COORDINATION PROCEEDING SPECIAL TITLE : (RULE 3,400) : 8 : SOCIAL MEDIA CASES : 9 :LEAD CASE :NO. FOR 10 THIS DOCUMENT RELATES TO: :FILING : PURPOSES 11 12 STATE OF TENNESSEE, ex rel, JONATHAN SKRMETTI, ATTORNEY GENERAL and REPORTER, :22STCV21355 : : : 13 V. : DEPOSITION UNDER VIDEO EXΑΜΙΝΑΤΙON OF VAISHNAVI JAYAKUMAR JANUARY 31, 2025 : 14 META PLATFORMS, INC., and INSTAGRAM, LLC : : 15 : CASE NO. 23-1364-IV : 16 17 18 19 20 21 22 23 VOLUME II Continued videotaped deposition of VAISHNAVI JAYAKUMAR, taken pursuant to notice, was held at the law offices of Baker Botts, LLP, 30 Rockefeller Plaza, New York, New York, beginning at a.m., on the above date, before Michelle L. Ridgway, a Registered Professional Reporter, Certified Court Reporter (NJ-CCR # XI02126), Certified Realtime Reporter, Certified Shorthand Reporter (CA-CSR # 14592), and Notary Public. 24 25 877-370-3377 Golkow Technologies, A Veritext Division Page 331 8:34 www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 147 of 211 Page 349 1 horrifying images of child sexual abuse and content 2 sexualizing minors? 3 MR. SNEED: Object to form. 4 THE WITNESS: Yes. 5 BY MR. WARREN: 6 Ω. And that content was actively 7 recommended by Instagram on various surfaces like 8 Explore and Discover People; is that fair? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes. 11 BY MR. WARREN: 12 Ω. And you agree those images were 13 14 15 inconsistent with Meta's publicly stated zero tolerance policy towards child sexual abuse material? 16 MR. SNEED: Object to form. 17 THE WITNESS: Yes. 18 BY MR. WARREN: 19 Ω. Okay. And that brings us to where 20 we left off yesterday, which is Exhibit 34. 21 Do you have that in front of you? 22 A. Yes. 23 Ω. Okay. And the first e-mail that we 24 looked at was and it was sent on 25 November 16, 2020. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 148 of 211 Page 350 1 Do you recall that? 2 A. Yes. 3 Ω. And Ms. sent an e-mail from 4 the Norwegian Broadcasting Corporation, correct? 5 A. Yes. 6 Ω. And you testified that you had 7 8 you recalled seeing that article at the time and being upset about it; is that right? 9 A. Yes, I was. 10 Ω. Okay. So we had been looking at 11 's response on November 16, 2020. Can 12 you direct your attention to that? 13 A. Yes. 14 Ω. And Mr. ends his response 15 16 by saying, "Clearly, we've handed over too much decision power to a technology that's not ready for 17 that yet, and it's beginning to erode people's 18 trust in our systems." 19 Did I read that correctly? 20 A. Yes. 21 Ω. Did you agree with Mr. 's 22 sentiment? 23 A. I agreed that we needed to be more 24 25 thoughtful about how many of these decisions were going to be automated versus how many of them 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 149 of 211 Page 351 1 should continue to be reviewed by humans. 2 Ω. Okay. Let's look at earlier in his 3 e-mail. He says this, and I'm quoting: 4 "Here's what Mark said about 5 6 sending our reviewers home on a press call back in March." 7 Did I read that correctly? 8 A. Yes. 9 Ω. And the Mark in question is Mark 10 Zuckerberg, correct? 11 A. Yes. 12 Ω. And the reference to sending 13 14 reviewers home is a reference to Meta's work-from-home policy during COVID-19, correct? 15 MR. SNEED: Object to form. 16 THE WITNESS: Yes. 17 MR. WARREN: Mr. Sneed, please let 18 me finish my question before you object. 19 MR. SNEED: Sure. 20 BY MR. WARREN: 21 Ω. What follows appears to be a 22 quotation from Mark Zuckerberg, given in response 23 to a question from an NPR reporter. Do you agree? 24 MR. SNEED: Object to form and 25 foundation. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 150 of 211 Page 352 1 THE WITNESS: I agree. 2 BY MR. WARREN: 3 Ω. Mr. highlights some aspects 4 of that quote in red. 5 Do you see that? 6 MR. SNEED: Object to form. 7 THE WITNESS: Yes. 8 BY MR. WARREN: 9 Ω. I'd like to read the beginning, and 10 you can tell me if I read it correctly. 11 Mr. Zuckerberg said: 12 "We have moved, or are in the 13 14 15 16 17 process of moving, the most sensitive types of content that need to get extra attention over to full-time employees for the time being. So that's the work on suicide and self-injury prevention. It's the work on child exploitation, counter 18 19 20 counter-terrorism. examples. We're actually surging the number of people who are working on those things." Those are kind of the types of 21 Did I read that correctly? 22 A. Yes. 23 Ω. That's not actually what happened, 24 is it? 25 A. Meta did increase the number of 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 151 of 211 Page 353 1 full-time employees who were working on those 2 things. The issue that was pointing out is 3 4 that before the classifiers could proactively before the classifiers could before humans can 5 review the content, classifiers need to, you know, 6 7 8 review and send the content over to people to be reviewed. And the classifiers weren't doing that correctly. 9 Ω. Well, let's look at Mr. 's 10 response on the first page of the document. 11 Can you flip to that? 12 A. Yes. 13 Ω. And actually, let me withdraw that 14 question. 15 Do you see your response to 16 Mr. on November 18, 2020? 17 A. Yes. 18 Ω. And do you see the first bullet of 19 your response? 20 A. Yes. 21 Ω. Can you read that, please. 22 A. "The CO," community operations, 23 24 "human review capacity for CEI is at approximately 30 percent. CO Safety suspects this will continue 25 to be true for many more months to come for large 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 152 of 211 Page 354 1 pools of queues with less than 20 percent projected 2 supply for the near future." 3 Ω. Does reading that cause you to want 4 5 to revise the testimony you just provided about the amount of human reviewers that were available to 6 review sexual abuse material? 7 MR. SNEED: Object to form. 8 THE WITNESS: So what Mark was 9 referring to, from my understanding, was the issue 10 11 of moving some of this work, a lot of this work, from external vendors to full-time employees. 12 13 14 15 And in the process number of full-time employees working on the issue surged, but the human review capacity was still really low because there's not that many full-time and so the 16 employees that can review CEI. 17 BY MR. WARREN: 18 Ω. So I don't know if I'm completely 19 following your answer, Ms. Jayakumar. 20 A. Yeah. 21 Ω. Just help me understand. 22 A. Yes. 23 Ω. So Mr. Zuckerberg appears to have 24 said, we're surging the number of people who are 25 working on those things, and "things" includes 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 153 of 211 1 child exploitation. Page 355 That's what he told the press. 2 Do you agree with that? 3 A. I do. 4 Ω. Okay. And in your response to 5 Mr. 6 's e-mail, you point out that human review capacity for child sexual abuse material is 7 at approximately 30 percent of what it should be. 8 Do you agree with that? 9 A. Yes, I do. 10 Ω. So that's not a surge, is it? 11 MR. SNEED: Object to form. 12 THE WITNESS: So it's not a surge 13 in the total number of people, but it's still a 14 surge in the total number of full-time employees. 15 So typically, CEI was not 16 handled -- 17 the review of CEI content wouldn't be handled by full-time employees. It would be 18 19 handled by contractors or vendors that are part of big outsourcing operations. 20 Those folks so he what he was 21 saying is that we're moving the review of this 22 content from those folks to full-time employees. 23 So surging the number of people was 24 surging the number of full-time employees who were 25 working on this. But the overall capacity is still 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 154 of 211 Page 356 1 very low. That's true. 2 BY MR. WARREN: 3 Ω. Okay. I just want to make sure I'm 4 clear so we can get a clear record. 5 It appears that the number of 6 full-time reviewers for this material went up. 7 8 A. Ω. Yes. But the number of contract 9 reviewers to review this material went down? 10 A. Yes. 11 Ω. Okay. And on net, that resulted in 12 less human review of child sexual abuse material 13 during COVID; is that correct? 14 MR. SNEED: Object to form. 15 THE WITNESS: Yes. Yes. 16 BY MR. WARREN: 17 Ω. And what Mr. Zuckerberg says, 18 again, is, "We're actually surging the number of 19 people who are working on those things." 20 Do you see that? 21 MR. SNEED: Object to form. 22 THE WITNESS: Yes. And I think 23 that that sounded a little bit more vague than it 24 should have. Because it sounded like it was the 25 total number of reviewers, when it sounds like what 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 155 of 211 Page 357 1 he really meant was the number of full-time 2 employees. 3 BY MR. WARREN: 4 Ω. Okay. Well, regardless of what he 5 meant or not, he said, "We're surging the number of 6 7 people who are working on child exploitation," correct? 8 MR. SNEED: Object to form. 9 THE WITNESS: Yes. 10 BY MR. WARREN: 11 Ω. But, in fact, Meta was not surging 12 the number of people working on child exploitation, 13 correct? 14 MR. SNEED: Object to form. 15 THE WITNESS: Not the total number 16 of people, no. 17 BY MR. WARREN: 18 Ω. Okay. And as a consequence of 19 that, the capacity to review potential child sexual 20 21 abuse material on Instagram went down significantly during COVID; isn't that correct? 22 MR. SNEED: Object to form. 23 THE WITNESS: Yes, that's correct. 24 BY MR. WARREN: 25 Ω. Okay. And you provided a stat to 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 156 of 211 Page 358 1 2 3 your teammates indicating that actually the projected supply of reviewers was less than 20 percent of what it needed to be; is that 4 correct? 5 A. Yes, correct. 6 Ω. Okay. And at no point did Meta 7 8 9 warn parents that, actually, Meta only had 20 percent of the bodies it would need to review potential child sexual abuse material on Instagram, 10 did it? 11 MR. SNEED: Object to form. 12 Foundation. 13 THE WITNESS: Not to my knowledge, 14 no. 15 BY MR. WARREN: 16 Ω. And do you think parents had a 17 right to understand that? 18 MR. SNEED: Object to form. Calls 19 for a legal conclusion. 20 THE WITNESS: I do think parents 21 had a right to know that. 22 BY MR. WARREN: 23 Ω. Okay. Especially during the 24 25 pandemic when more kids were necessarily online; is that right? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 157 of 211 Page 371 1 problem had not been fixed in the intervening 2 12 months? 3 A. Yes. 4 Ω. Okay. You can put that document to 5 one side. 6 I'm going to hand you what's been 7 marked for identification as Exhibit 36. This is a 8 document bearing the Bates stamp 9 ΜΕΤΑ3047MDL-020-00151907. 10 (Document marked for identification 11 as Jayakumar Exhibit 36.) 12 BY MR. WARREN: 13 Ω. And I'll represent this was 14 produced to us by Meta and found within your 15 electronic files. Do you have any reason to 16 dispute that? 17 A. No. 18 2. Okay. Do you recall seeing this 19 document before? 20 A. I'm familiar with this 21 interstitial, yes. 22 Ω. Okay. And am I correct that this 23 is a screen shot of Instagram with the word 24 "childporn" written in the search bar? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 158 of 211 Page 372 And this displays an interstitial 1 Ω. 2 that pops up in response to entering that search, 3 correct? 4 MR. SNEED: Object to form. 5 THE WITNESS: Yes. 6 BY MR. WARREN: 7 Ω. Okay. Can you read the text of 8 this interstitial, please. 9 A. "This May Be Associated With Child 10 Sexual Abuse. 11 "This term is sometimes associated 12 with sexual images or videos of children. 13 "Child sexual abuse or viewing 14 sexual imagery of children can lead to imprisonment 15 and other severe personal consequences. 16 "This abuse causes extreme harm to 17 children and searching and viewing such materials 18 adds to that harm. 19 "If you see this type of activity 20 here, see how to report it to us and to law 21 enforcement in the link below." 22 Ω. Okay. And then there are two 23 options underneath that text. 24 Do you see that? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 159 of 211 Page 373 1 Ω. The first option says, "See How To 2 Report," correct? 3 A. Yes. 4 Ω. And then the second option says, 5 "See results anyway." 6 Do you see that? 7 A. Yes. 8 Ω. So that option allowed the user to 9 10 11 click past the warning and see the results of the child porn search despite the warning; is that correct? 12 MR. SNEED: Object to form. 13 THE WITNESS: Yes. 14 BY MR. WARREN: 15 Ω. And the interstitial makes clear 16 17 18 19 20 21 22 23 24 25 that what may appear after clicking that button include child sexual abuse and sexual imagery of children, correct? MR. SNEED: Object to form. THE WITNESS: Yes. THE VIDEOGRAPHER: Excuse me, Counsel. I'm sorry to interrupt. I'm getting a notification that their audio is not okay. So can we just go off the record just to clear that up? MR. WARREN: No, we can't. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 160 of 211 Page 377 1 Wall Street Journal article in that post, correct? 2 A. Yes. 3 Ω. I'm going to hand you a copy of 4 that, which we'll mark for identification as 5 Exhibit 38. 6 (Document marked for identification 7 as Jayakumar Exhibit 38.) 8 BY MR. WARREN: 9 Ω. Is this the article that you 10 referenced in your LinkedIn post? 11 A. Yes. 12 Ω. The lead reporter is someone named 13 Jeff Horwitz; is that right? 14 A. Yes. 15 Ω. Did you speak to him? 16 A. I did. 17 Ω. When? 18 A. I don't know the exact date, but 19 when he was preparing this article. 20 Ω. Okay. And the article is dated 21 November 27, 2023; is that right? 22 A. Yes. 23 Ω. Okay. What did you tell 24 Mr. Horwitz? 25 A. I said that we either need to ramp Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 161 of 211 Page 378 1 up our content detection capabilities or we should 2 3 Ω. just block out minor content from Recommendations. Okay. Can you please read the 4 first few paragraphs of the article to the jury. 5 A. "Instagram's Reels video service is 6 7 8 designed to show users streams of short videos on topics the system decides will interest them, such as sports, fashion or humor. 9 "The Meta Platforms-owned social 10 11 12 app does the same thing for users its algorithm decides might have a prurient interest in children, testing by The Wall Street Journal showed." 13 Ω. Keep going, please. 14 A. "The Journal sought to determine 15 16 17 18 what Instagram's Reels algorithm would recommend to test accounts set up to follow only young gymnasts, cheerleaders and other teen and preteen influencers active on the platform. 19 "Instagram's system served jarring 20 21 doses of salacious content to those test accounts, including risqué footage of children as well as 22 overtly sexual adult videos and ads for some of 23 the biggest U.S. brands." 24 Ω. Okay. Thank you. 25 877-370-3377 Ms. Jayakumar, all of these Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 162 of 211 Page 379 problems were ones that you had flagged internally 1 2 as early as March 2020, just weeks after you 3 4 started at Meta, correct? MR. SNEED: Object to form. 5 THE WITNESS: Many of these 6 problems I had flagged as early as March. I think 7 from March to May 2020. 8 BY MR. WARREN: 9 Ω. And you agree that these problems 10 11 are inconsistent with Meta's publicly stated zero-tolerance policy for child exploitation 12 material, correct? 13 MR. SNEED: Object to form. 14 THE WITNESS: Yes. I agree. 15 BY MR. WARREN: 16 Ω. All right. Let me direct your 17 attention to the top of the second page. 18 Do you see a quote from a Meta vice 19 president named Samantha Stetson? 20 A. Yes. 21 Ω. Could you please read that to the 22 jury. 23 A. "'Our systems are effective at 24 reducing harmful content, and we've invested 25 billions in safety, security and brand suitability 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 163 of 211 Page 380 1 " 2 solutions, said Samantha Stetson, a Meta vice president who handles relations with the 3 4 advertising industry." Ω. Okay. Based on your three and a 5 6 7 8 half years working at Meta on child safety issues, do you think it's accurate that Meta's systems are effective at reducing harmful content related to child sexualization and child exploitation? 9 MR. SNEED: Object to form. 10 THE WITNESS: Without any measure 11 of how much of a reduction we are talking about, 12 the statement is technically correct. 13 BY MR. WARREN: 14 Ω. Okay. Well, let's unpack that, 15 then. 16 What do you think constitutes 17 effectiveness when it comes to reducing harmful 18 content related to child sexualization and child 19 exploitation? 20 A. 21 I don't think we can talk about effectively reducing harmful content without 22 agreeing more broadly on what those what those 23 24 25 metrics are. Is it okay for 1 percent of the content on the platform to include child sexual abuse material, 2 percent, zero percent? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 164 of 211 Page 381 As far as I know, I don't think 1 2 3 Meta has ever specified what percentage would be considered effective. 4 And so, yes, the systems reduce 5 harmful content, but it's not but not, in my 6 opinion, by as much as they should. 7 Ω. Meta has specified a percentage, 8 hasn't it? It said zero. It has a zero-tolerance 9 policy for this material, isn't that correct? 10 A. It is 11 MR. SNEED: Object to form. 12 Characterization. 13 THE WITNESS: 14 Sorry. It has a zero percent tolerance for 15 child sexual exploitation. It looks like Sam 16 17 Stetson was talking about reducing harmful content more broadly. 18 BY MR. WARREN: 19 Ω. Right. And I'm asking you about 20 21 reducing harmful content related to child sexual exploitation. That's my specific question. So I'm 22 going to ask you again. 23 Based on your experience working at 24 Meta for three and a half years on child safety 25 issues, do you think it's accurate to say that 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 165 of 211 Page 382 1 Meta's systems are effective at reducing harmful 2 content related to child sexualization and child 3 exploitation? 4 MR. SNEED: Object to form. 5 THE WITNESS: No. 6 BY MR. WARREN: 7 Ω. Thank you. 8 So please flip to the next page. 9 Near the bottom, do you see the 10 11 paragraph beginning with "Current and former Meta employees"? 12 A. Yes. 13 Ω. Could you please read that to the 14 jury. 15 A. 16 17 18 19 20 21 22 23 Ω. 24 "Current and former Meta employees said in interviews that the tendency of Instagram algorithms to aggregate child sexualization content from across its platform was known internally to be a problem. Once Instagram pigeonholes a user as interested in any particular subject matter, they said, its recommendation systems are trained to push more related content to them." Do you agree with the proposition in the first sentence there? 25 A. 877-370-3377 I do. Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 166 of 211 1 2 3 Ω. Page 383 Were you one of the former Meta employees who said that to The Wall Street Journal? THE WITNESS: Mike? 4 Can I consult with my lawyer? 5 MR. WARREN: Of course. Why don't 6 7 we go off the record and the tech can fix the audio too. 8 THE VIDEOGRAPHER: The time right 9 now is 9:12 a.m. We are off the record. 10 (Short break.) 11 THE VIDEOGRAPHER: The time right 12 now is 9:15 a.m. We're back on the record. 13 BY MR. WARREN: 14 Ω. I just want to direct your 15 attention to one more part of this article, 16 Ms. Jayakumar. 17 Near the bottom of that page it 18 says, "Preventing the system from pushing noxious 19 content to users interested in it" let me 20 withdraw the question. 21 Okay. Near the bottom of that 22 page, Ms. Jayakumar, it says: 23 "Preventing the system from pushing 24 25 noxious content to users interested in it requires significant changes to the recommendation Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 167 of 211 Page 384 algorithms that also drive engagement for normal 1 2 users." 3 Did I read that correctly? 4 A. Yes. 5 Ω. Okay. And do you agree with that 6 proposition? 7 A. Yes. 8 Ω. Okay. And then the next sentence 9 10 11 12 13 in the article says, "Company documents reviewed by the Journal show that the company's safety staffers are broadly barred from making changes to the platform that might reduce daily active users by any measurable amount." 14 Did I read that correctly? 15 A. Yes. 16 Ω. And based on your three and a half 17 18 19 20 years of working at Instagram, do you agree that its safety staffers are broadly barred from making changes to the platform that might reduce daily active users by any measurable amount? 21 22 MR. SNEED: Object to form. THE WITNESS: I'm not familiar with 23 24 25 are the company documents reviewed here. But it is generally quite difficult to make changes that it is generally pretty difficult to make 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 168 of 211 Page 385 1 safety changes that might impact growth or daily 2 active users by any significant amount. 3 BY MR. WARREN: 4 Ω. And do you believe that's the 5 reason why Instagram failed, over three years, to 6 solve the problem of the Reels algorithm 7 8 recommending child sexualization content to pedophiles? 9 MR. SNEED: Object to form. 10 THE WITNESS: I think it's a 11 significant reason. 12 BY MR. WARREN: 13 Ω. Okay. Turning to the next page 14 here, it says here, "Even before the 2020 launch of 15 16 Reels, Meta employees understood that the product posed safety concerns, according to former Meta 17 employees." 18 Do you see that? 19 A. Yes. 20 Ω. Do you agree with that proposition? 21 A. Yes. 22 Ω. And finally, skip down one 23 paragraph to the paragraph beginning with "In." 24 Do you see that? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 169 of 211 Page 449 1 Speculation. 2 THE WITNESS: Not based on what 3 this document said in 2021. 4 BY MR. KOSSLYN: 5 Ω. Okay. I'm going to turn us over to 6 Exhibit 45, which is META3047MDL-014-00206538. 7 (Document marked for identification 8 as Jayakumar Exhibit 45.) 9 BY MR. KOSSLYN: 10 Ω. So this appears to be an e-mail 11 thread between you and and 12 is that correct? 13 A. Yes. And Antigone Davis. 14 Ω. Yes. 15 I think you may have referenced 16 this yesterday, but just refresh my recollection, 17 who is ? 18 A. 19 policy team with was on the safety and me. She reported into 20 Antigone. 21 Ω. 22 Safety State of Play And you see the header, "Re: Child Timely/Action needed"? 23 A. Yes. 24 Ω. 25 Do you think this was discussing the Child Safety State of Play document we were 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 170 of 211 Page 450 1 just referencing? 2 A. That's right. Yeah. 3 Ω. Okay. Take as long as you need to 4 review this. 5 I'll note that I'm mostly going to 6 be focusing on the original e-mail here from 7 the 24th of September. And specifically 8 9 starting on the fourth page, which is Bates ending in 541, in the section labeled, "Not ready; work 10 has not yet been prioritized," and then the rest of 11 the document after that. 12 A. Yeah. 13 Ω. So I'm going to walk through the 14 15 various cells in this table and just ask you some follow-up questions. 16 I think we've gone over some of 17 these features before, but I just want to confirm 18 that they they are referring to the same things 19 that you testified on yesterday. 20 A. Sure. 21 Ω. And provide some additional 22 information. 23 So, first, "Aggregated content on 24 IG surfaces," can you explain what this was. 25 A. Sure. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 171 of 211 Page 451 1 So our 2 3 4 5 6 7 8 the classifiers that were in place were able to proactively detect individual pieces of content that may violate some of our policies, for example, around child exploitation, suicide, self-injury, and so on. But they were not able to proactively detect and take action on groups of content that would appear in one spot together. 9 10 11 12 13 14 15 So the Explore tab, for example, on Instagram will have many pieces of content that all appear at the same time, or Hashtag page might be the same, Reels certainly is like that. And the classifiers that were currently in place could not detect when those aggregated surfaces were violating any of our policies. 16 Ω. Now, all of these cells are under 17 the header, "Not ready; work has not been 18 prioritized and more resources are needed"; is that 19 right? 20 A. Yes. 21 Ω. Do you know how many resources 22 would have been needed to build this product? 23 MR. SNEED: Object to form. 24 THE WITNESS: I would not be able 25 to say that definitively. I would need to see I Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 172 of 211 Page 452 1 would need to, you know, really build out a product 2 scope and then make that call. 3 BY MR. KOSSLYN: 4 Ω. Okay. Why would this product have 5 6 been useful to the platform and improve safety for minors? 7 A. Because the because places like 8 Explore, Hashtag pages, Reels, all feature - 9 10 11 featured minors and were viewable by minors. improving our ability to detect and remove problematic content from them would have made them So 12 13 safer, both as creators as well as viewers of the platform. 14 Ω. Thank you. 15 And then the next cell reads, 16 "Block reach of unconnected adults to IG minors." 17 Can you describe what that product 18 would have been? 19 A. Yeah. So we, I think, talked about 20 21 22 23 this earlier in the deposition, but I was really keen on making sure that minors could not be proactively contacted by adults who had no connection to them. 24 This is something that was already 25 in place as protection on Facebook Messenger, and 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 173 of 211 Page 453 1 it wasn't in place on Instagram. However, we have 2 far more discoverability of minors on Instagram 3 4 5 than on Facebook. There are also just a lot more young people using Instagram than, say, Facebook or Facebook Messenger. 6 So this was a gap that I had 7 identified, saying that we, you know, don't have 8 9 10 11 12 13 so Interop refers to interoperability between Instagram DMs and Facebook Messenger. you have probably seen by now that you can kind of switch between those accounts when you're sending messages to one another. this protection in place on Instagram. The product team So 14 And for that launch, product was 15 16 not agreeing to protect this rule for cross-network messages sent by Facebook adults to Instagram 17 18 minors, which would have amplified the problem even more, from my perspective. 19 Ω. Why did product not agree to do 20 that, as you understand? 21 A. From what I recall, and from this 22 23 document, it sounds like this was actually ability of Facebook adults to connect with the 24 Instagram minors was a big growth bet for the 25 company in some, you know, really understandable 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 174 of 211 Page 454 1 ways. 2 For example, parents who might have 3 an account on Facebook but not Instagram, wanting 4 5 to reach out to their kids who are on Instagram but maybe not on Facebook. 6 And so this was a big growth bet 7 for interoperability. 8 Ω. And was the lack of resources 9 10 A. 11 raised as a reason why Facebook was not doing this? No. This wasn't due to a lack of resources. This was a disagreement on the value of 12 blocking the reach of unconnected adults. 13 Ω. Okay. And the next cell, do you 14 see, "IIC classifier use on IG"? 15 A. Yes. 16 Ω. I think we also discussed this 17 yesterday. IIC stands for inappropriate 18 interactions with children; is that right? 19 A. Yes. 20 Ω. Can you just briefly describe what 21 this feature was or would have been. 22 A. The inappropriate interactions with 23 children classifier there are actually a number 24 of IIC classifiers. Some of them looked at the 25 direct messages, the kind of messages that, you 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 175 of 211 Page 455 1 know, were going back and forth. Others looked at 2 3 the way in which adults were trying to reach out to minors. 4 But overall, the premise was that 5 these -- this set of classifiers were monitoring 6 7 and proactively detecting any inappropriate interactions so they could either protect the minor 8 more or penalize the adult more, or some 9 combination of both. 10 Ω. And do you see in the cell to the 11 right where it says, "We need about a week or two 12 of CO review capacity in order to calibrate the 13 classifier for IG use"? 14 A. Yes. 15 Ω. Was that your understanding of the 16 only requirement that would have been needed in 17 order to make this technology work? 18 A. So the classifier, as it stood at 19 20 this time, was trained on Facebook data and not really Instagram data, so that's what we mean by 21 calibrating it. 22 Yes, I think that sounds about 23 right. It wouldn't have taken too much review 24 capacity to get it calibrated for Instagram use. 25 Ω. Do you see below that where it 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 176 of 211 Page 456 1 says, "This is low hanging fruit and fully 2 supported by Child Safety CI and IG Policy. We 3 just need help getting CO to prioritize this"? 4 A. Yes. 5 Ω. What is your understanding of why 6 CO was not prioritizing this? 7 MR. SNEED: Object to form and 8 foundation. 9 THE WITNESS: I don't remember why 10 CO wasn't prioritizing this back then. This was 11 this was, you know, nearly five years ago. 12 But wait. This was in 2020? 13 BY MR. KOSSLYN: 14 Ω. Mm-hmm. 15 A. So, yeah, I don't remember I 16 don't remember exactly why. I don't have the 17 documentation. 18 But in 2020, CO was extremely 19 20 strapped in its resourcing abilities because of COVID-19. 21 Ω. Understood. 22 Do you see below that, "Upsell 23 reporting for Disappearing Mode"? 24 A. Yes. 25 Ω. Can you explain what that was? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 177 of 211 Page 457 1 A. 2 3 4 5 So Disappearing Mode, which is also known as Vanish Mode on Instagram today, allows you to send messages to people and then have it disappear, you know, right after. They are ephemeral messages. 6 And so the retention period 7 within on the back end for Instagram was 8 14 days. It was one hour for Messenger. 9 Since Disappearing Mode is an 10 11 12 13 14 15 16 17 18 ephemeral form of communication, in order to protect minors who might be subject to inappropriate messages or grooming messages, we wanted to upsell and really promote the value of reporting within Disappearing Mode. And so the ask that we had made was support reporting upsells within Disappearing Mode to educate the user that even ephemeral messages can be reported and that they have a limited window of time in which to do 19 SO. 20 Ω. 21 22 23 24 In the cell to the right it says, "However, product does not have the CO resource capacity to support reporting upsells within Disappearing Mode to educate the user that even ephemeral messages can be reported." 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 178 of 211 Page 458 1 Ω. Can you explain what that means? 2 A. It meant that product did not - 3 well in terms 4 5 6 7 8 9 10 from what I recall, it's think what this is referring to is that if we upsold reporting, and if that resulted in a large volume of reports from, say, minors, we wouldn't have the capacity to review those reports, and the team that would review the reports would be the community operations team, and we wouldn't have the resources to do that. - I 11 Ω. Okay. So if minors were getting a 12 13 14 lot of ephemeral messages that contained reportable content, then they would not know to report those. And the concern is if they did know to report 15 those, that there would be too many reports? 16 MR. SNEED: Object to form. 17 THE WITNESS: Yes, that's right. 18 BY MR. KOSSLYN: 19 Ω. Okay. Do you know, did Meta 20 21 disclose to the public at all that there were, on ephemeral messages, a large volume of IIC or CEI 22 being conveyed? 23 MR. SNEED: Object to form. 24 THE WITNESS: I don't agree that a 25 large amount of IIC or CEI was being conveyed 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 179 of 211 Page 459 1 through disappearing messages. That that may 2 have been the case, but I don't know if that's 3 true. 4 But, no, Meta didn't provide any 5 6 external warning that, you know, that there may be a likelihood of IIC or CEI taking place within 7 disappearing messages. 8 BY MR. KOSSLYN: 9 Ω. Okay. And then turning to the next 10 cell that says, "Sex trafficking classifier." 11 Can you explain what this was. 12 A. This would be a classifier to 13 proactively detect content that is promoting or 14 soliciting the trafficking of sex workers or, 15 actually sorry trafficking of humans for sex, 16 not necessarily sex workers. 17 Ω. Okay. And can you read what the 18 cell on the right says, just the first sentence. 19 A. "We have one engineer dedicated to 20 21 22 23 this work and we have no data science support to do some basic understand work to lower the strikes threshold for prostitution and sexual solicitation strikes (currently set at 17x, which is too high)." 24 Ω. What does that mean, do you know? 25 A. So we had an engineer that was Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 180 of 211 1 2 3 4 5 6 Page 460 willing to build the classifier, but we needed data science to actually help us understand what the behaviors looked like, like what are the types of behaviors that might constitute sex trafficking on the platform, so that we could build and train the classifier with datasets. Without doing that, it was very difficult to justify lowering the strike threshold. 7 8 9 10 11 12 13 17X, by any measure across the industry, is a very, very high strike threshold. That means that you could incur 16 violations for prostitution and sexual solicitation, and upon the 17th violation, your account would be suspended. 14 Ω. Okay. So there could be someone 15 16 who did it 15 times on the platform and there would be no action against them? 17 MR. SNEED: Object to form. 18 19 20 THE WITNESS: Their content would be removed, you know, with the strike. But their account would be allowed to remain on the platform, 21 yes. 22 BY MR. KOSSLYN: 23 Ω. In your understanding, was Meta 24 reliably accurate at detecting prostitution 25 solicitations? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 181 of 211 1 Page 461 MR. SNEED: Object to form. 2 THE WITNESS: Not to my 3 recollection. 4 BY MR. KOSSLYN: 5 Ω. Okay. Do you know if Meta ever 6 7 8 9 warned the public, or disclosed to the public, that it was not capable of reliably, or with a high degree of reliability, determining when someone is soliciting prostitution? 10 MR. SNEED: Object to form. 11 THE WITNESS: No, I don't think so. 12 BY MR. KOSSLYN: 13 Ω. Okay. Scrolling down slightly to 14 the section this is now outside the table 15 "Recommendation Age Assurance." 16 A. Okay. 17 Ω. I'm going to walk through just a 18 19 few sentences throughout this. I think this is mostly covered in the Exhibit 6 document we 20 previously went through. 21 But do you see where it says, "We 22 need leadership support to push product to act on 23 knowledge we already have, i.e., clear our backlog, 24 disable verified U13s"? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 182 of 211 Page 462 1 Ω. Why do you understand that you 2 needed leadership support to do that? 3 A. All these 4 all these decisions, all these moves, would require significant product 5 input and sorry significant product effort. 6 And so we needed this was a call, I think, to 7 8 our leadership to persuade the product team to provide those resources and support this work. 9 Ω. And who would your leadership have 10 been? 11 A. This Child Safety State of Play 12 would have gone to Meta's content policy 13 leadership. 14 Ω. Do you remember who that would have 15 been at this point? 16 A. I don't know exactly who Antigone 17 sent it to. I didn't have visibility into that. 18 Ω. And then "disable verified U13s." 19 Is this just referring to the 20 linked accounts issue or is there something else 21 this incorporates as well? 22 A. I think this refers to disabling 23 any account that we know is under the age of 13. 24 Ω. Were there other accounts that Meta 25 knew were under the age of 13 that weren't being 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 183 of 211 Page 463 1 disabled, that you're aware of? 2 A. I don't recall. 3 Ω. Do you know who might? 4 A. Antigone might. 5 Ω. And, "Against this backdrop, our 6 enforcement is not consistent against all apps for 7 a single known person." 8 Is this all referring to the 9 cross-linking? 10 MR. SNEED: Object to form. 11 THE WITNESS: That's right. Yes. 12 BY MR. KOSSLYN: 13 Ω. 14 15 16 Do you know, was it just the Facebook and Instagram problem that was described in Exhibit 6 earlier? Was it also Instagram to Facebook or some other app to other app? 17 A. I don't recall - I don't recall 18 how fully like, which range of behaviors that 19 was describing. 20 Ω. 21 A. Okay. Do you know who would? Probably the identity integrity 22 team. 23 Ω. And then near the end of this page, 24 "We also need leadership to support the basic 25 principle that if we're using a signal to predict 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 184 of 211 Page 464 1 age for business purposes, it should be used to 2 enforce on age." 3 What did you understand this to be 4 referring to? 5 This is the same page. It's the 6 second-to-last sentence. 7 MR. SNEED: Object to form. 8 THE WITNESS: I'm trying to 9 understand the context. 10 I believe this is referring to the 11 12 13 14 15 fact that we sometimes use non-stated age signals, looking at, you know, different aspects of the account's behavior, to try and predict the age of that account, to decide what kind of ads might be most relevant to them, what the best type of 16 promoted content would be for them. And if we are 17 18 19 20 21 able to do that for business purposes, it says then, we should also be able to do that for integrity purposes. We should also be able to do that to identify an account, for example, as being under 13 and removing it from the platform. 22 BY MR. KOSSLYN: 23 Ω. So why would Meta use this 24 algorithmically inferred age rather than stated age 25 for these business purposes? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 185 of 211 Page 465 1 MR. SNEED: 2 3 4 5 6 7 8 9 10 Object to form. THE WITNESS: Sometimes because stated age doesn't exist. Instagram only started collecting stated age in 2019. So in those cases, you would need an algorithmically inferred age. There is also value to algorithmically inferred age sometimes. It's sometimes a more accurate predictor than stated age, especially in cases where people lie about their age. 11 BY MR. KOSSLYN: 12 Ω. So you had referred to some of 13 those business purposes as being advertising, 14 right? 15 A. Yes. 16 Ω. Do you know if there are other 17 business purposes? 18 A. That is the main one I can think 19 of. 20 Ω. Okay. Going down to the table now 21 22 in the next page. This is the cell marked, "Age Management" on the right. I think much of this is 23 24 the same as what ended up in Exhibit 6, so I'm not going to belabor it. 25 But just to refresh your Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 186 of 211 Page 466 1 recollection on the language. "In the meantime, 2 our basic COPPA compliance is at risk when the 3 4 product does not prioritize checkpointing and disabling U13s." 5 If you can go now down to the cell 6 beneath it, this is under, "Not ready; work has not 7 been prioritized and more resources are needed." 8 "Age data on IG." 9 Do you see that? 10 A. Yes. 11 Ω. Can you read the first two 12 sentences? 13 A. "Age models do not work as well in 14 15 16 17 IG because IG only started to collect age data in December 2019. In addition, Adam Mosseri will not support upselling age collection for existing users because he sees it as intrusive data collection." 18 2. Okay. So why would the fact that 19 20 Instagram had only started collecting age data in 2019 affect the quality of its age models? 21 A. You need to be able to provide data 22 23 points to the age models to train them, and if you don't have the stated age for example, if you 24 don't have the stated age of an account, you can't 25 really compare that against all the predictive 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 187 of 211 1 2 signals you have and understand how close or know, how close to correct or far from correct you Page 467 you 3 are. 4 Ω. And this means that even if Meta 5 6 7 wanted to use those age models for safety features, they would not have been as reliable as those on Facebook, for example? 8 MR. SNEED: Object to form. 9 10 11 12 13 THE WITNESS: Yes. Yes, that's right. I believe at this point it was 2020 actually, I think yeah, so this was in 2020, and so we only had stated age for maybe two-thirds of IG users. 14 BY MR. KOSSLYN: 15 Ω. Yeah. I think that language is in 16 the very next sentence. 17 So do you know or were you ever 18 informed why Instagram did not collect age data 19 from users prior to December 2019? 20 A. Instagram had very little 21 information on its users prior to in many ways, 22 for a long time. 23 You know, if you think about the 24 nature of Instagram as a product, you have a 25 username that's not actually your real name. 877-370-3377 Golkow Technologies, A Veritext Division You www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 188 of 211 Page 468 1 2 have a profile photo. It doesn't have to be your profile photo. 3 For a long time it wasn't 4 necessarily seen as something where you needed to 5 assign, like, a whole, like, identity to it. 6 But then in 2019, I think IG 7 decided to actually start collecting this data. 8 Ω. Do you know why they decided to do 9 so in 2019? 10 A. It predates my time there. 11 Ω. Okay. And then as you just read, 12 Adam Mosseri allegedly did not support age 13 collection on existing users. 14 Do you understand the basis for 15 that objection? 16 MR. SNEED: Object to form and 17 foundation. 18 THE WITNESS: I understand what 19 he's saying here, yes. 20 BY MR. KOSSLYN: 21 Ω. Just to clarify. Do you understand 22 that this is Adam Mosseri's language or this is 23 someone characterizing Adam Mosseri? 24 MR. SNEED: Object to form. 25 877-370-3377 THE WITNESS: This is someone Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 189 of 211 Page 469 1 characterizing Adam Mosseri. 2 BY MR. KOSSLYN: 3 Ω. Okay. Do you understand if 4 Instagram eventually did start collecting age 5 information from existing users? 6 A. Yes, we did. 7 Ω. Do you know why Instagram started 8 doing so? 9 A. I think it took a lot of ongoing 10 conversations with Instagram leadership. 11 Eventually, you know, we just let's see. I want 12 to be careful about that. 13 Eventually, the company decided to 14 15 16 17 18 2. collect age. I know it was an ongoing conversation at the Adam level. But I don't think I can speculate on the exact reason why the company decided to eventually collect age for everyone. Okay. And it's the very last 19 20 21 22 23 24 sentence, you "recommend engaging with IG leadership on collecting age for the remaining roughly 33 percent and prioritizing age model maturity on Instagram." Do you understand that some of the remaining 33 percent may have been under the age of 25 13 at this time? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 190 of 211 Page 470 1 A. They may have been. 2 Ω. Okay. And Instagram would have no 3 way of knowing, because they did not disclose their 4 age one way or another to Instagram; is that right? 5 A. Yes. 6 MR. SNEED: Object to form. 7 BY MR. KOSSLYN: 8 Ω. Okay. Do you know if, 9 10 subsequently, when Instagram did require users to state their age, even if they had an Instagram 11 account, if any of those users said they were under 12 the age of 13? 13 A. If they had said they were under 14 15 the age of 13, they wouldn't have been allowed to remain on the platform or to sign up for an 16 account. So yeah so I don't know. 17 Ω. I mean, to your knowledge, were 18 19 there any users that, in fact, were not allowed to continue to have their account because they had 20 indicated they were 12 or younger? 21 A. Not to my knowledge. Not to my 22 direct knowledge. 23 Ω. Okay. Would you have had direct 24 knowledge if that was the case? 25 A. 877-370-3377 I wouldn't necessarily have had Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 191 of 211 Page 471 1 direct knowledge. That would have gone through 2 3 our, you know, moderators, and it would have wouldn't have necessarily been raised to me. - it 4 5 Because it was such a automated process it was such a process, an not an automated process, but 6 it was such a process. 7 Ω. I understand. 8 If you can please scroll up. This 9 is now to your reply to this original e-mail. This 10 will be on Page 2, I believe. Page 1 and 2. 11 You say, "Hi, Karina, thank you for 12 sharing suggested edits from me are below." 13 And then one of your proposed edits 14 is "Add to 'Not Ready'. 15 "We do not enforce against certain 16 concerning behaviors at scale across 17 Facebook/Instagram." 18 Do you see that? 19 A. Yes. 20 Ω. Can you describe what you're 21 describing here? 22 A. So there were there are accounts 23 on Instagram that don't have too much activity 24 within the account, you know, when it comes to 25 posts or stories that they are sharing. But all 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 192 of 211 Page 472 1 2 they do is leave very sexual, you know, quote, unquote, romantic messages on pictures of minors. 3 And that's all the account does. 4 5 That, to me, was a very clear indicator of an account that was dedicated to minor 6 sexualization. And this was something that we were 7 really only able to address upon escalation. 8 And escalation is a long process. 9 10 It means that someone may report the account. The account may be found to not be be not violating, 11 12 13 14 15 and then some third party, maybe, that they have access to is able to escalate it to us, or to someone within the company, and then we would review it again, at which point it could be disabled. 16 Ω. If you can scroll all the way to 17 18 the top. This is on the first page. When you say, "Hi, Sarah on 1) This is a policy gap." 19 Can you describe what you mean here 20 by a policy gap? 21 A. Yes. So Sarah was saying in her 22 e-mail that this might be because our AI tooling is 23 not mature enough to proactively detect this 24 content at scale. But what I was saying is that we 25 don't have a policy around what to do with accounts 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 193 of 211 Page 473 1 2 that are only dedicated to leaving sexualizing comments on minors' posts. 3 And so I was saying that this isn't 4 a question you can't have a product tool without 5 the foundational policy. So I was saying this is 6 not a product issue, this is a policy issue. 7 Ω. So how would this have been 8 enforced at this point, if there were such 9 accounts? 10 A. Yeah. So at this point, what would 11 happen is if -- let's say I, a user on Instagram, 12 reported this piece of content, it would have not 13 violated any of our this account sorry it 14 15 16 would not have violated any of our policies. would have gotten back a message saying there was no violation. I 17 I then would have had to find some 18 19 20 21 way to escalate it, perhaps by contacting an Instagram employee or, you know, finding a civil society group that has a relationship with the company. 22 23 24 25 And then those employees or those external groups would escalate it internally to their contacts at the company who would then, you know, send it to an operations team, saying, hey, 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 194 of 211 Page 474 1 can you take a look at this case again. So that's 2 what an escalation would have looked like. 3 Ω. 4 In your experience, how many teenagers have relations with Instagram employees 5 or civil society groups? 6 MR. SNEED: Object to form and 7 foundation. 8 THE WITNESS: In my experience, 9 very few. 10 BY MR. KOSSLYN: 11 Ω. Okay. And just to confirm on this, 12 13 14 this is something you're proposing to add to "not ready," as in Instagram, at present, was not doing anything to address this problem? 15 MR. SNEED: Object to form. 16 THE WITNESS: Yes. That's right. 17 BY MR. KOSSLYN: 18 Ω. Do you know did Instagram or Meta 19 warn the public that this was a risk and that this 20 was not covered conduct? 21 MR. SNEED: Object to form. 22 THE WITNESS: No. Meta did not. 23 BY MR. KOSSLYN: 24 Ω. And to your knowledge, did Meta 25 revise its policies to address this at some point? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 195 of 211 Page 475 1 A. I don't remember. 2 Ω. Okay. I'm done with this document. 3 We can set it aside. 4 Sorry. If we can reverse that for 5 6 a moment and go back to Bates ending in 543. this would be Page 6, 7 MR. SNEED: 8 The e-mail or the MR. KOSSLYN: No. This is no, 9 this is the table. 10 BY MR. KOSSLYN: 11 Ω. This is under the "Global use of 12 age models." 13 A. Okay. 14 Ω. Can you read the first sentence 15 there. 16 A. "The best performing age model is 17 18 19 20 21 22 23 24 25 not yet globally operational, which leads to a patchwork of integrity measures across the company that leverage age. We need more review resources to establish ground truth data on age in priority marks. We should start by engaging with Core Dimensions (the product team who own the age models) on what markets we've identified and stack ranked as a Policy org to prioritize with their available H2 resources." 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 196 of 211 Page 476 1 Ω. Just to confirm, your understanding 2 3 was the vast performing age model was not available or operational on Instagram itself; is that right? 4 A. The best performing age 5 Ω. 6 Sorry. Let me withdraw that. Your understanding was that the age 7 8 model that was available in the United States for Instagram was not highly performing; is that right? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes, that's correct. 11 BY MR. KOSSLYN: 12 Ω. Okay. And is your understanding 13 14 15 that that complicated or reduced the ability of integrity measures to be employed for minors on the platform? 16 MR. SNEED: Object to form. 17 THE WITNESS: Yes, that's correct. 18 BY MR. KOSSLYN: 19 Ω. Okay. Thank you. You can turn the 20 document aside. 21 So I just would like us to briefly 22 return to Exhibit 9, just to refresh your 23 recollection on the date. 24 This document is from March 2021; 25 is that correct? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 197 of 211 Page 728 1 A. It's right here. 2 Ω. Okay. Do you see that? 3 A. Yes. 4 Ω. Now, this was presented to you by 5 defense counsel as a list of all the great things 6 that Instagram has done to keep things safe. 7 Is that a fair characterization of 8 the line of questions you were presented with? 9 MR. SNEED: Object - object to 10 form. 11 THE WITNESS: Yes. 12 BY MR. WARREN: 13 Ω. So I want to go through and discuss 14 some of these with you, and we're going to start 15 with the various kinds of nudges on Instagram. 16 Are you familiar with the word, 17 "nudges"? 18 A. Yes. 19 Ω. And nudges either prompt users to 20 21 consider alternative behavior or provide them with resources on a particularly harmful issue; is that 22 fair? 23 A. Could you direct me to the page I 24 should be looking at? 25 Ω. 877-370-3377 Sure. Why don't we take a look at Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 198 of 211 Page 729 1 an example. So the second page, at the top, 2 July 2019. 3 4 5 Do you see that? A. July 2019. Yes. Ω. Okay. And that's labeled, "Comment 6 Warnings, correct? " 7 8 A. Ω. Yes. And comment warnings "prompted 9 people to reconsider posting comments that may be 10 hurtful"; is that correct? 11 A. Yes. 12 Ω. Okay. Now, a comment warning 13 14 15 wouldn't actually prevent Instagram users from posting hateful comments on someone else's post, correct? 16 17 A. Ω. Correct. 18 19 20 So would you consider this a temporary Band-Aid or a foundational change in Instagram's architecture, geared towards child safety? 21 MR. SNEED: Object to form. 22 THE WITNESS: I think this is a 23 valuable this is a valuable tool for Instagram 24 25 to have launched, in part, because the context makes a really big difference. And the same term, 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 199 of 211 Page 730 1 2 3 the same expression, said by different people, to different audiences, could or could not be hateful. And so I think including comment 4 warnings was a good way 5 6 7 a good nudge. We also know that when users are nudged or reminded that their policies might violate the rules, they have a higher likelihood of, you know, retracting what 8 they want us to do retracting what they are 9 going to do. 10 BY MR. WARREN: 11 Ω. Right. So fair to say that it was 12 a helpful intervention? 13 A. Yes. 14 Ω. 15 Okay. But it was not foundationally going to prevent hateful comments 16 and bullying and harassment on the platform, right? 17 A. 18 19 I would disagree with that. think that nudges are a really effective way of discouraging these types of posts from happening. I 20 So I think it is an effective foundational move. 21 Ω. All right. Very well. Let's talk 22 about a different one. 23 In March 2021, Meta says on this 24 document that it "restricted adults over 18 from 25 starting private chats with teens they are not 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 200 of 211 Page 731 1 connected to on Instagram and Messenger." 2 Do you see that? 3 A. 4 Ω. Yes. Okay. Now, that tool's efficacy 5 relies on the accuracy of Instagram's understanding 6 of the age of users, correct? 7 A. Yes. 8 Ω. If Instagram is wrong about how old 9 10 either the teenager or the adult is, then this restriction is not going to accomplish very much; 11 would you agree? 12 MR. SNEED: Object to form. 13 THE WITNESS: I agree. 14 BY MR. WARREN: 15 Ω. Okay. And, in fact, as of 16 March 2021, I believe you testified that 17 Instagram's age inference model was not able to 18 accurately identify many adults who described 19 themselves as teenagers; is that correct? 20 A. Correct. 21 Ω. And conversely, Instagram's age 22 inference tool was not able to identify many 23 teenagers who identified themselves as adults, 24 correct? 25 A. That's correct. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 21 of 211 Page 63 1 several weeks after the launch of Reels in the 2 United States, the issue still hadn't been 3 addressed? 4 A. Yes. That's fair. 5 Ω. Okay. Now, please flip to the page 6 ending in 37249. 7 Do you see the section called, 8 "Immediate Product Vulnerabilities"? 9 A. Yeah. 10 Ω. And there's a bullet that says IG, 11 meaning Instagram, correct? 12 A. Mm-hmm, yes. 13 Ω. Okay. Can you read the first 14 sub-bullet. 15 A. "Lack classifiers to detect and 16 take action on problematic aggregated content on 17 surfaces like Explore, hashtag pages, Reels." 18 Ω. Okay. Now we've talked about that 19 problem with respect to Reels already, correct? 20 A. Yes. 21 Ω. Can you explain that problem with 22 respect to Explore and hashtag pages? 23 24 A. First of all, what is Explore? The Explore Explore is a tab on 25 Instagram that you can tap into that shows you a 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 22 of 211 Page 64 1 2 3 4 5 range of images and videos, some might be from people you follow, many might be from people you don't follow. It's tailored to your interests based on a number of factors, including what you've seen before on Instagram. 6 Ω. 7 A. Okay. And what are hashtag pages? Hashtag pages are, if you type in a 8 9 hashtag, say #puppies, you'll get an Instagram page that's full of content from people you may know or 10 may not know, that's that has all been tagged 11 with that hashtag, #puppies. 12 Ω. 13 14 Okay. So how do you understand the phrase "lack classifiers to detect and take action on problematic aggregated content on surfaces like 15 Explore and hashtag pages"? 16 A. I think when I wrote this, or when 17 18 19 20 I contributed this input to the team that wrote this, I was referring to the fact that individual pieces of content can be taken down or taken action on, but when pieces of content that might be 21 borderline, appearing in aggregate, don't have any 22 solution for them. 23 So the best way I can think of 24 explaining that is through an example. 25 If someone were to post a picture 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 23 of 211 Page 65 1 of a minor that was nude, that would be a violation 2 of our child sexual abuse materials policy and that 3 piece of content would be taken down. 4 If somebody had an account that was 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 entirely dedicated to teen gymnasts in skimpy clothing, no individual piece of content, it's fine to post team gymnasts on Instagram, there's many reasons, gym clubs have their own accounts, but if the whole account is dedicated to that, in aggregate that looks like a lot of content that is sexualizing minors. The same applies if you, say, type in the #teen, and then have a whole page of results showing teens in, you know, very little clothing. And so while individual pieces of content may be okay on their own and not have been posted with any bad intention, in aggregate, they promote minor sexualization and we need a way to deal with that. We need a way to either filter them out of the hashtag pages, make sure that only a certain percentage of those non-violating images appear in Explore or hashtag pages. And we certainly need to filter them out of services like Reels. 25 Ω. Okay. And so the classifiers that 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 24 of 211 1 Page 66 Instagram had at that time actually were not able 2 3 to detect and take action with respect to that issue on Explore and hashtag pages, correct? 4 A. Yes, correct. 5 Ω. And they weren't able to take 6 7 8 action with respect to that issue on Reels for the additional reason that they weren't just tuned to dealing with short-form video at all, correct? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes, correct. 11 BY MR. WARREN: 12 Ω. Okay. Did Meta leadership warn 13 parents or kids about any of these risks that you 14 just testified about? 15 MR. SNEED: Object to form and 16 foundation. 17 THE WITNESS: Meta did not warn 18 19 20 parents and kids about the specific types of risks within the product. But they did have digital literacy programs and media literacy programs where 21 22 they try to tell parents how to, you know, govern their children's use of the platform. 23 BY MR. WARREN: 24 Ω. Right. And helpful though that is, 25 those materials don't actually inform parents or 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 25 of 211 Page 67 1 kids that Instagram didn't have an automated way to 2 3 detect and take action on problematic aggregated content on Explore, Reels, and hashtag pages, 4 right? 5 MR. SNEED: Object to form. 6 THE WITNESS: Yes, that's correct. 7 MR. WARREN: Okay. It's been about 8 an hour, maybe more. Why don't we take a break. 9 Okay, thanks. 10 THE VIDEOGRAPHER: The time right 11 now is 10:19 a.m. We are off the record. 12 (Short break.) 13 THE VIDEOGRAPHER: The time right 14 now is 10:35 a.m. We're back on the record. 15 BY MR. WARREN: 16 Ω. Okay. Ms. Jayakumar, before the 17 break, you had testified about a problem with Reels 18 related to the visibility of young people to 19 adults; is that correct? 20 A. Yes. 21 Ω. Okay. Now, it's correct that 22 23 around this time, Meta was considering a new safety intervention called Private By Default, is that 24 true? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 26 of 211 Page 68 1 Ω. 2 MR. SNEED: Okay. And that Object to form. Sorry. 3 Go ahead. 4 BY MR. WARREN: 5 Ω. Okay. That intervention would have 6 limited the visibility of young people to adult 7 strangers? 8 A. Yes. 9 Ω. Okay. So I'm going to hand you 10 another document. This will be Exhibit 7. 11 (Document marked for identification 12 as Jayakumar Exhibit 7.) 13 BY MR. WARREN: 14 Ω. And for the record this is 15 ΜΕΤΑ3047MDL-050-00285759. 16 Okay. Ms. Jayakumar, is this a 17 chat thread between you and from 18 August 24, 2020? 19 A. Yes. 20 Ω. And can you please tell the jury 21 who Ms. is. 22 A. was my counterpart 23 on the privacy team. So she looked after privacy 24 policy. 25 Ω. 877-370-3377 Okay. And can you please read the Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 27 of 211 Page 69 1 first two messages in this chat? 2 A. From : "Hi! Hope you're 3 doing well. Do you happen to know the status of 4 the IG Smart Defaults launch?" 5 And then from me: "I do we were 6 7 8 9 10 11 Ω. 12 13 14 15 initially supposed to launch private by default on September 2nd after approval from IG leads, but that has now been scrapped due to growth concerns, and the product team has been given some time to come up with ways to mitigate the drop in growth." Okay. Now when you said the launch of private by default was "scrapped due to growth concerns," did those concerns include the safety feature's potential impact on growing the overall number of Instagram users? 16 A. Yes. 17 MR. SNEED: Object to form. 18 BY MR. WARREN: 19 Ω. Did the concerns include the 20 feature's potential impact on the amount of time 21 that users spend on Instagram? 22 MR. SNEED: Object to form. 23 THE WITNESS: I don't think it was 24 about time spent. 25 I think it was about how people discover one another and how the number of total 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 28 of 211 Page 70 1 users increases. 2 BY MR. WARREN: 3 Ω. Fair enough. 4 So did the growth concerns then 5 6 include the potential safety feature's possible impact on engagement between Instagram users? 7 A. 8 Yes. MR. SNEED: Object to form and 9 foundation. 10 BY MR. WARREN: 11 Ω. Okay. And those growth impacts 12 would have meant a decrease in Instagram's revenue, 13 correct? 14 MR. SNEED: Same objections. 15 THE WITNESS: Yes. It would have. 16 BY MR. WARREN: 17 Ω. Okay. And that was a concern that 18 resulted in pushing back the launch date of the 19 safety feature, right? 20 MR. SNEED: Object to form and 21 foundation. 22 THE WITNESS: Yes. 23 BY MR. WARREN: 24 Ω. Okay. Is it fair to say this is an 25 example of Meta leadership prioritizing growth over 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 29 of 211 Page 71 1 safety? 2 MR. SNEED: Object to form and 3 foundation. 4 THE WITNESS: Yes. 5 BY MR. WARREN: 6 Ω. Okay. Now was a version of private 7 by default ultimately launched? 8 A. Yes. 9 Ω. Okay. When it launched, did it 10 apply to all teen accounts? 11 A. No. It applied to new accounts 12 that were signing up to the platform that were 13 declaring themselves to be teenagers. 14 Ω. Okay. Was it ever applied to all 15 teen accounts? 16 A. It was applied to all teen accounts 17 at the end of last year, as part of the teen 18 accounts launch. 19 Ω. Okay. So that would be the end of 20 2024, correct? 21 A. Yes. 22 Ω. Okay. So fully four years after 23 this chat with Ms. ? 24 A. Yes. 25 Ω. Okay. So I'm going to ask you some Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 30 of 211 Page 72 1 questions about how private by default worked. But 2 first a preliminary question: 3 What is a default setting? 4 A. A default setting is something that 5 your account comes with when you sign up for it. 6 It means that for example, a default setting 7 8 9 might look like, when you sign up for an account, nobody can message you. You can go in and change that setting if you want. But as default, nobody 10 can come nobody can message you. 11 Ω. Okay. And you say that you can go 12 in and change the setting if you want. Do 13 Instagram users generally do that? 14 A. Generally speaking, most users, not 15 16 17 just within Instagram, don't change their default settings until they are in a moment of crisis. And then they look to perhaps adjust their settings. 18 Ω. Okay. So the name "private by 19 20 default" suggests this was a default. Would you agree? 21 22 23 24 A. Yes. Ω. Was it actually a default setting? MR. SNEED: Object to form. THE WITNESS: It was I don't 25 know if it would have been considered a default 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 201 of 211 Page 732 1 Ω. Okay. 2 3 So while this sounded like a very good intervention, there were some significant limitations to it at the time it launched, correct? 4 MR. SNEED: Object to form. 5 THE WITNESS: Yes. I think this is 6 a good example of some of an intervention that 7 is a helpful tool but doesn't address the 8 foundational issue of not being clear about user 9 age on the platform. 10 BY MR. WARREN: 11 Ω. Okay. So would this one be closer 12 to a temporary Band-Aid or a foundational product 13 solution? 14 MR. SNEED: Object to form. 15 THE WITNESS: I would say it was 16 more of a Band-Aid. 17 BY MR. WARREN: 18 Ω. Okay. And then let's look at right 19 above right above there. 20 Do you see, February 2021, "We 21 launched expert-backed resources." 22 Do you see that? 23 A. Yes. 24 Ω. Okay. Well, expert-backed 25 resources do not prevent the platform from 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 202 of 211 Page 746 1 counsel about that? 2 A. Yes. 3 Ω. And I believe you testified that 4 the version that launched didn't impact growth as 5 much as the product we wanted. 6 Did you say something to that 7 effect? 8 A. Yes. 9 Ω. And what did you mean by that? 10 A. The original version of private by 11 12 13 default that was scheduled to launch on, I think, September 2nd was scrapped from product because of growth concerns. And so the version that 14 eventually did launch resulted in a drop of growth, 15 16 I believe, from the documents, of 2 to 5 percent, and the product team had built a bunch of 17 mitigations to ensure that that drop in growth had 18 been mitigated as much as possible. 19 Ω. Okay. And the version that 20 launched of private by default actually, let me 21 withdraw the question. 22 Can you direct your attention to 23 Exhibit 50, please. 24 A. Yes. 25 Ω. And can you direct your attention Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 203 of 211 Page 747 1 to the page ending in 16780. 2 A. Yes. 3 Ω. Okay. And do you see the section 4 called, "Prior Mitigations"? 5 A. Yes. 6 Ω. And so is it correct that in order 7 8 to offset the negative impact on growth metrics, Instagram came up with new notifications and 9 reminders to launch at the same time that it was 10 rolling out a version of private by default? 11 A. That's correct. 12 Ω. Okay. And that included contact 13 join notifications? 14 A. Yes. 15 Ω. And follow request reminder? 16 A. Yes. 17 Ω. 18 19 20 Okay. So the mitigations were mitigations targeted at ensuring that Instagram didn't actually lose out on any growth when it launched the safety feature; is that correct? 21 22 23 24 MR. SNEED: Object to form. THE WITNESS: It ensured that Instagram had the least negative impact on growth. There was some negative impact on the growth. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 204 of 211 Page 748 Okay. And what about the version 1 Ω. 2 3 of private by default that launched was less safety protected than the version that you thought should 4 be launched? 5 A. 6 7 A number of things. But most critically, private by default was originally supposed to be for all users, and, instead, it was 8 9 for new users and it was for new users under the age of 16. 10 Ω. Okay. Not existing users? 11 A. No. 12 Ω. And not new or existing users under 13 the age of 18, correct? 14 A. Correct. 15 Ω. Can you flip the page of this 16 exhibit. 17 A. I'm sorry. Not under the age of 18 18, unless they were in EU or the UK. 19 Ω. Okay. Very well. 20 But as far as American kids were 21 concerned, no one under the age of 18 was getting 22 this? 23 A. No. 24 Ω. Okay. I'm sorry. Let me withdraw 25 the question because I misstated. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 205 of 211 1 Page 749 No one ages 17 and 18 was getting 2 this, correct? 3 A. Correct. 4 Ω. Okay. So are you on the next page, 5 16781? 6 A. Yes. 7 Ω. Can you direct your attention to 8 the top right corner? 9 A. Yes. 10 Ω. Okay. Now, yesterday you talked 11 about how private by default wasn't actually a 12 default setting. It was a preselection; is that 13 right? 14 A. In my opinion, yes, it's not a true 15 default setting. 16 Ω. Okay. And is the right-hand screen 17 shot reflective of that testimony? 18 A. Yes. 19 Ω. Okay. Can you walk us through 20 that. 21 A. 22 So when a user signs up for the platform, so a new teen user, they say their 23 birthday, they state their birthday. And if that 24 birthday is below 16 so there's no age 25 verification. But if they say they are under the 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 206 of 211 Page 750 1 age of 16, then they would get this screen 2 immediately after asking them if they want to have 3 a private account or a public account. And private 4 was selected by default. 5 Ω. Okay. And in your view, that's not 6 a default setting in actuality, correct? 7 MR. SNEED: Object to form. 8 THE WITNESS: That, to me, is not a 9 true default setting, no. 10 BY MR. WARREN: 11 Ω. All right. Let's turn our 12 attention to the topic of child sexual abuse 13 material, which came up again in counseling from 14 the defense questioning from the defense lawyer. 15 Do you recall that? 16 A. Yes. 17 Ω. And let's look at Exhibit 51. 18 A. Yes. 19 Ω. Do you recall Mr. Sneed questioning 20 about Exhibit 51? 21 A. Yes. 22 Ω. Okay. And the exhibit concerns 23 something called profile-level reporting for CEI on 24 Instagram; is that correct? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 207 of 211 Page 757 1 THE WITNESS: Correct. 2 BY MR. WARREN: 3 Ω. And why did you have that caveat in 4 your answers to Mr. Sneed? 5 A. I wanted to be clear that this 6 7 8 9 wasn't my opinion, and that's not why I shared it. I shared it because this was a study from the Cyber Policy Center. Ω. 10 A. Okay. So that wasn't your opinion? This wasn't my opinion. 11 Ω. Okay. Why wasn't it your opinion? 12 A. Well, I hadn't done I hadn't 13 14 researched whether or not Instagram truly had the most aggressive approach across all the platforms 15 16 17 Ω. at this moment in time. So I don't know if I would have agreed with this study or this statement. Let's actually examine how 18 aggressive Instagram's approach to suicide and 19 self-injury aggregated content was. 20 And let's look at Exhibit 68, which 21 was presented to you by the defense lawyer. 22 Do you have that in front of you? 23 A. Did you say 68? 24 Ω. I did. 25 877-370-3377 Now, unfortunately, this was Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 208 of 211 Page 758 1 2 3 4 5 produced in a format without page numbers, so I'm going to do my best to orient you. About halfway through the deck, there is a slide called, "Approach to SSI Across Meta." And it's got a table with a bunch of Xs. 6 A. Yes. 7 Ω. Okay. Now does this table indicate 8 whether these approaches applied to one, some, or 9 all of Meta's family of apps? 10 A. This would suggest that this 11 applies across the family of apps. 12 Ω. Okay. But let's take a look a few 13 pages forward from that. The slide entitled, "What 14 Did We Learn About SSI and Teens?" 15 Do you see that? 16 A. Yes. 17 Ω. Okay. And do you see the first box 18 in the table under the column "Learning"? 19 A. Yes. 20 Ω. And can you read that to the jury, 21 please. 22 A. 23 "Significant gaps, complexity and inconsistency in usability, access and content of 24 the resources we offer to users." 25 Ω. Okay. And then product work for 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 209 of 211 Page 759 1 that is identified as, "In progress and not 2 roadmap," depending on the sub-bullet, correct? 3 A. Correct. 4 Ω. Okay. And there are other things 5 that are identified as not roadmapped on this 6 slide, correct? 7 A. Correct. 8 Ω. And one of those is: "There are 9 additional levers we could pull to address 10 11 problematic SSI content exposure, e.g., content dilution." 12 Correct? 13 A. Correct. 14 Ω. All right. And this site indicates 15 that teens see more suicide and self-injury content 16 than adults, correct? 17 A. Yes. 18 Ω. 19 And while that was roadmapped to be addressed in the second half of that year, meaning 20 2023, it hadn't actually been addressed as of that 21 point in time, correct? 22 A. Yes. 23 Ω. Okay. So in 2023, 11 years after 24 the acquisition of Instagram by Meta, it had not 25 addressed the fact that teens were presented with 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 210 of 211 Page 760 1 more suicide and self-injury content than adults? 2 MR. SNEED: Object to form. 3 BY MR. WARREN: 4 Ω. Is that correct? 5 A. Yes. 6 Ω. Okay. And it had not addressed the 7 8 9 10 fact that there was significant gaps, complexity, and inconsistency in the usability, access, and content of the SSI-related resources that Meta offered to users, correct? 11 MR. SNEED: Object to form. 12 THE WITNESS: Yes. 13 BY MR. WARREN: 14 Ω. And it had not pulled all the 15 levers that it could to address problematic SSI 16 content exposure to kids, right? 17 MR. SNEED: Object to form and 18 foundation. 19 THE WITNESS: Yes. 20 BY MR. WARREN: 21 Ω All right. Now, you were asked 22 some questions about various things that you said 23 to external stakeholders. 24 Do you remember that? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com 1 2 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 CERTIFICATE Page 211 of 211 Page 770 3 4 5 6 I HEREBY CERTIFY that the witness was duly sworn by me and that the deposition is a true record of the testimony given by the witness. It was requested before completion 7 of the deposition that the witness, VAISHNAVI JAYAKUMAR, have the opportunity to read and sign the deposition transcript. 8 9 Michelle L. Gray 10 MICHELLE L. RIDGWAY, 11 Certified Shorthand Reporter, (CA-CSR # 14592) 12 (Certified Court Reporter (NJ-CCR # XI02126) 13 Registered Professional Reporter, Certified Realtime Reporter 14 and Notary Public Dated: February 4, 2025 15 16 17 18 19 20 21 22 23 24 25 (The foregoing certification of this transcript does not apply to any reproduction of the same by any means, unless under the direct control and/or supervision of the certifying reporter.) 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 31 of 211 Page 73 1 setting in layman terms, but it was a default 2 preselected option on a screen. 3 BY MR. WARREN: 4 Ω. What do you mean by "default 5 preselected option"? 6 A. Sorry for how confusing that 7 sounds. 8 So a typical if you said to 9 somebody, oh, this is an account that is private by 10 11 12 default, I think an assumption would be, I sign up, I'm a teenager on the platform, my account is private. I will go into settings and change that 13 to be a public account if I want to. 14 The way in which private by default 16 17 18 19 15 ultimately launched was, I sign up, I say I'm a teenager, I get a screen asking me if I want to have a public account or a private account, and the private option is selected. So on that screen, I make a choice to say, well, this is a public 20 account or a private account. Yeah. 21 Ω. Okay. So it's not a baked-in 22 default that just operates from the jump. ump. Rather, 23 it's a screen that gives you an option private or 24 public? 25 877-370-3377 MR. SNEED: Object to form. Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 32 of 211 Page 74 1 THE WITNESS: Correct. 2 BY MR. WARREN: 3 Ω. Okay. If private by default had 4 5 6 7 been an actual default of the kind you were just describing, with the option to disable it in settings, do you think that would have resulted in more teens with private accounts? 8 MR. SNEED: Object to form and 9 foundation. 10 THE WITNESS: Yes. I think so. 11 BY MR. WARREN: 12 Ω. And why do you think that? 13 A. We have data to suggest that teens 14 generally don't change their settings. 15 Even with the screen option, which 16 17 18 19 is, you know, a pretty easy toggle between two choices, six out of ten teens said that they weren't changing their settings. They were just going to go with whatever the app suggested. 20 So if it had been truly a true 21 22 default, that they had to go into their settings later and change, it's highly unlikely that most of 23 them would have done that. 24 Ω. Okay. So that ultimately would 25 have been more protective of teen safety, do you 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 33 of 211 Page 75 1 think? 2 A. Yes. I think so. 3 Ω. Okay. But it's not what Instagram 4 did, right? 5 A. It's not what Instagram did. 6 Ω. Okay. And is it fair to say that 7 8 9 growth concerns are what motivated Instagram to pick the less safety protected version of private by default? 10 MR. SNEED: Object to form and 11 foundation. 12 THE WITNESS: This is the yes, 13 this was the conversation in 2020. 14 BY MR. WARREN: 15 Ω. 16 A. So that was a yes to my question? Yes, yes. 17 Ω. Okay. Let's change topics. I want 18 19 20 to ask you some general questions about Instagram's business model and its user base. Are you familiar with those topics? 21 A. Yes. 22 Ω. Okay. Instagram is an ad-supported 23 platform, correct? 24 A. Yes. 25 Ω. And that means it earns revenue by Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 34 of 211 Page 77 1 THE WITNESS: Yes. 2 BY MR. WARREN: 3 Ω. How is it able to do that? 4 MR. SNEED: Object to form and 5 foundation. 6 THE WITNESS: The Instagram 7 8 9 looks at a number of signals about the user to understand what types of content might be most appealing to them. Both organic content posted by 10 11 other users, as well as promoted content, so things like ads. 12 13 14 15 16 17 18 That can range from signals such as accounts that they follow, accounts that their friends follow, content that they've liked or engaged with in some way. There's a wide variety of signals, but those collectively give the app a good idea of what a user might be interested in seeing. 19 BY MR. WARREN: 20 Ω. Okay. And when you use the word 21 22 "signals," just so that the jury understands, does that refer to data about users? 23 24 A. Ω. Yes. 25 Okay. And you said there's a wide variety of data about users that Instagram collects 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 35 of 211 Page 78 1 in order to target ads, right? 2 A. Yes. 3 Ω. Okay. And I'd like to understand 4 some types of that data. 5 Does that include data about users' 6 interests? 7 MR. SNEED: Object to form and 8 foundation. 9 THE WITNESS: Yes. 10 BY MR. WARREN: 11 Ω. Does it include photos and videos 12 of the user? 13 MR. SNEED: Object to form and 14 foundation. 15 THE WITNESS: Yes. 16 BY MR. WARREN: 17 Ω. Does it include geolocation 18 information about the user? 19 MR. SNEED: Object to form and 20 foundation. 21 THE WITNESS: Yes. 22 BY MR. WARREN: 23 Ω. Does it include the usernames of 24 the user? 25 MR. SNEED: Object to form and 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 36 of 211 Page 79 1 foundation. 2 THE WITNESS: Yes. 3 BY MR. WARREN: 4 Ω. Does it include the gender of the 5 user? 6 MR. SNEED: Same objections. 7 THE WITNESS: Yes. 8 BY MR. WARREN: 9 Ω. Okay. Are there any other kinds of 10 signals that you can describe sitting here today? 11 MR. SNEED: Objection. 12 THE WITNESS: They could look at a 13 14 15 16 17 18 signal such as interests that you and a friend have in common, or interests that a large number of your friends have in common, and for example, infer that that might suggest something about your own interests as somebody who is friends with all these different accounts. 19 BY MR. WARREN: 20 Ω. Is one of the signals that 21 Instagram analyzes, "hover time," meaning how long 22 a user hovers over a certain image? 23 MR. SNEED: Object to form and 24 foundation. 25 THE WITNESS: I don't know with 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 37 of 211 Page 85 1 THE WITNESS: Yes. That's right. 2 BY MR. WARREN: 3 Ω. Okay. So I'm going to hand you 4 what has been marked as Exhibit 8. 5 (Document marked for identification 6 as Jayakumar Exhibit 8.) 7 BY MR. WARREN: 8 Ω. And for the record, this is Meta 9 A. Oh, I would like to change my 10 earlier answer. 11 Ω. Sure. 12 A. Could you repeat the question, the 13 previous question. 14 Ω. Up to 2021, is it correct that 15 Instagram was just taking users at their word with 16 respect to their age? 17 A. There were some measures Instagram 18 19 20 was taking to be more certain about age. So for example, it did have a reporting flow where you could report users for being underage and they 21 would review that. 22 If an account was reported for 23 anything for any other type of violation, the 24 reviewer would also review the account and make 25 sure that it, you know, wasn't actually an underage 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 38 of 211 Page 86 1 person on the platform. So there were a couple of 2 other things the company was doing. 3 MR. WARREN: Okay. Can we take a 4 quick break? 5 THE WITNESS: Mm-hmm. 6 THE VIDEOGRAPHER: The time right 7 now is 10:51 a.m. We are off the record. 8 (Short break.) 9 THE VIDEOGRAPHER: The time right 10 now is 10:54 a.m. We're back on the record. 11 BY MR. WARREN: 12 Ω. Okay. So, Ms. Jayakumar, you said 13 there were some measures Instagram was taking to be 14 more certain about age, right? 15 A. Yes. 16 Ω. Okay. And you said it had a 17 reporting flow where people could report users for 18 being underage, correct? 19 A. Yes. 20 Ω. 21 22 Now, that wouldn't actually address the problem of an adult pretending to be underage in order to engage in sexual exploitation, right? 23 MR. SNEED: Object to form. 24 THE WITNESS: Yes. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 39 of 211 Page 87 1 Ω. 2 3 Okay. And you said that another mitigation measure was that an account reported for another type of violation would be reviewed and 4 analyzed for age, right? 5 A. Yes. 6 Ω. So in the situation that you just 7 8 described of a 23-year-old man sexually grooming a 15-year-old, something bad would have to happen 9 first in order for the 20-year-old - 23-year-old 10 account to get flagged in the first place, right? 11 MR. SNEED: Object to form. 12 THE WITNESS: Yes. That's right. 13 BY MR. WARREN: 14 Ω. Okay. And then after that had 15 happened, the account would be analyzed for their 16 age and potentially action would be taken, right? 17 MR. SNEED: Object to form and 18 foundation. 19 THE WITNESS: Yes. 20 BY MR. WARREN: 21 Ω. Okay. So were those mitigation 22 measures you just described sufficient to prevent 23 the grooming hypothetical that you gave moments 24 ago? 25 A. No. They are not sufficient for Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 40 of 211 Page 88 1 that. 2 But I was answering the question 3 4 of, you know, did Meta have anything in place to identify underage users. 5 Ω. Fair enough. 6 So it had some things in place, but 7 8 not enough to prevent some very serious safety issues for kids? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes. 11 BY MR. WARREN: 12 Ω. Okay. Let me direct your 13 attention, actually, back to what was previously 14 marked as Exhibit 6. 15 And this would be the child 16 safety/state of play from September 24, 2020. 17 Just let me know when you're there. 18 A. Yep, I'm here. 19 Ω. Can you please direct your 20 attention to the page ending in 37246. 21 Do you see the section labeled "U13 22 Enforcement"? 23 A. Yes. 24 Ω. What does U13 enforcement mean? 25 A. It means removing users under the Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 41 of 211 Page 89 1 age of 13 from the platform. 2 Ω. Okay. Now, in the middle of this 3 description, it reads: 4 There's also a current U13 backlog. 5 Do you see that? 6 A. Yes. 7 Ω. And what does it say in the 8 parentheses next to that? 9 A. "Approximately 450,000 job 10 backlog." 11 Ω. Okay. And what do you understand 12 that to mean? 13 MR. SNEED: Object to foundation. 14 THE WITNESS: There are 450,000 15 reports or noted incidents of potential users under 16 the age of 13 that need to be reviewed and 17 addressed. 18 BY MR. WARREN: 19 Ω. Okay. And a little bit further 20 down, it says: 21 This backlog is currently not 22 23 resourced and has been escalated to Ops Resources. Do you see that? 24 A. Yes. 25 Ω. 877-370-3377 Okay. What does that mean? Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 42 of 211 Page 90 1 MR. SNEED: 2 Object to foundation. THE WITNESS: It means that there 3 weren't 4 5 there weren't staff allocated to reviewing and addressing the backlog, like reviewing those cases. And so we have escalated 6 the case to the ops resources team which might have 7 8 head count and ability to staff backlog. to review the 9 BY MR. WARREN: 10 Ω. Okay. Let me direct your attention 11 to what we'll mark as Exhibit 8. 12 (Document marked for identification 13 as Jayakumar Exhibit 9.) 14 MR. WARREN: For the record, this 15 is a document ending in I'm sorry. This is a 16 document bearing the Bates number 17 ΜΕΤΑ3047MDL-020-00278850. 18 BY MR. WARREN: 19 Ω. Let me know when you've had a 20 chance to review that. And take your time. 21 Are you ready? 22 A. Yes. 23 Ω. Okay. This appears to be a chat 24 thread among you, and 25 from March 23, 2021. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 43 of 211 Page 91 1 Do you agree with that? 2 A. Yes. 3 Ω. And was this chat thread conducted 4 in the ordinary course of your work in Meta? 5 A. Yes. 6 Ω. Can you explain to the jury who the 7 other folks on this chat are? 8 A. and were our safety 9 10 And then policy representatives in the Asia-Pacific region. was my child safety counterpart who 11 worked on all the other products outside of 12 Instagram. 13 Ω. And that would include Messenger 14 and Facebook? 15 A. Yes. 16 Ω. Okay. So the chat starts with 17 Ms. She writes to you, Ms. and 18 Ms. Seems to indicate that some of Meta's 19 partners had been asking for in-app reporting for 20 underage accounts. 21 Do you agree with that? 22 A. Yes. 23 Ω. What does "in-app reporting" mean? 24 A. So at the time, if you wanted to 25 report a user for being under 13, you had to go to 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 44 of 211 Page 92 1 2 3 a web form. And so, you know, go out of the app essentially, go to a form and then report that account for being under 13. 4 She was asking whether we could 5 6 7 8 9 10 just include it within the app itself. So within the app, if you click on a piece of content now, you can report it for a variety of violations. was asking if we could also include an option to report the account within the app itself for being underage. She 11 Ω. Would that have facilitated more 12 reports of under-13 Instagram users? 13 A. Yes. It's a much faster, much more 14 seamless way to report users. 15 Ω. So is it fair to say that prior to 16 March 23, 2021, that in-app reporting option did 17 not exist on Instagram for the reporting of 18 underage users? 19 A. Yes. That's accurate. 20 Ω. Okay. Now, you respond to 21 Ms. What was your response? 22 A. I said that I think we would 23 have do you want me to read it out or? 24 Ω. 25 Well, you can read it out, but I also want to know what you were thinking. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 45 of 211 1 2 A. Page 93 I think I said yes, I think folks would very much be interested in doing this. But 3 4 it would be a significant lift from the legal and privacy perspective. 5 Ω. Okay. Why did you think it would 6 7 A. 8 9 10 be a significant lift from a legal perspective? I think in-app reporting just increases the volume of accounts that need to be that would that would come in our way. And we already had that huge backlog of 450,000 accounts, - -- 11 which I think you saw in a previous report. 12 And so it would increase the legal 13 14 risk if we're now getting all these additional reports of potentially underage users. 15 Ω. I see. So the concern was that the 16 17 18 19 20 in-app reporting option would actually work and result in a large number of reports of potential under-13 users, but then Instagram wouldn't have the resources to actually go through all of those; is that fair? 21 MR. SNEED: Object to form and the 22 characterization. 23 THE WITNESS: That was my that 24 was my perspective and that was my impression from 25 working with the different teams across the board. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 46 of 211 Page 94 1 BY MR. WARREN: 2 Ω. Okay. And you're obviously not a 3 lawyer, correct? 4 A. Yes. 5 Ω. 6 7 So you were addressing this and raising this concern more from a resource standpoint; is that right? 8 9 THE WITNESS: Yes. MR. SNEED: Object to form. My -- my point 10 was that and I 11 12 you know, I fully support including in-app reporting for under-13 accounts. I was talking to more about the likelihood of 13 14 us being able to get that passed, rather than the inherent value of getting it passed. 15 BY MR. WARREN: 16 Ω. I see. 17 So your concern then seems to be 18 19 20 twofold. One, if we do this, are we going to have the bodies to actually go through the reports. That's one you already talked about, right? 21 22 A. Ω. Mm-hmm, yes. And then it seems like you had a 23 second concern which was, I'm not sure we're going 24 to be able to convince people to do this; is that 25 correct? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 47 of 211 Page 95 1 MR. SNEED: Object to form. 2 Characterization. 3 THE WITNESS: I think my primary 4 5 6 7 8 9 concern is that I don't think we can convince people to do this. And one of the reasons is because it would introduce an enormous risk to have all of these additional accounts that we know we're not getting resourcing for. We already know that we have 450,000 accounts that we haven't been able 10 to staff and review. Increasing that by some 11 incredibly large number, which is I think would 12 13 be off the table for our legal and privacy teams from a risk perspective. 14 BY MR. WARREN: 15 Ω. Understood. 16 Okay. Now I would like to just 17 direct you to Ms. 's response. She says: 18 We're working on improving our 19 20 underage reporting flows and backlog. It is a longer term project though (clearing the backlog is 21 the immediate concern right now). 22 Did I read that correctly? 23 A. Yes, you did. 24 Ω. So she was agreeing with you there 25 still was a large backlog of reports of potential 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 48 of 211 Page 96 1 under-13-age users, right? 2 A. Yes. 3 Ω. Okay. 4 MR. SNEED: Object to form. Sorry. 5 BY MR. WARREN: 6 Ω. Okay. And then she provides what 7 she calls a data point. I'm going to quote here. 8 She says: 9 For reporting, currently there's no 10 11 way for users to report under-13 users on Facebook in-app and it takes over seven screens for users to 12 report under-13 users on Instagram. 13 Did I read that correctly? 14 A. Yes. 15 Ω. Do you have any reason to doubt the 16 accuracy of what said? 17 A. No. was very well-plugged 18 into our child safety work. 19 Ω. Okay. And then she goes on to say: 20 On enforcement, we currently have a 21 backlog of over 2.5 million cases age-related 22 reports to review. 23 Do you see that? 24 A. Yes. 25 Ω. 877-370-3377 Do you have any reason to doubt the Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 49 of 211 Page 97 1 accuracy of that? 2 A. No, I don't think that was I 3 think that was very accurate. 4 And it kind of backs up an earlier 5 6 7 8 point that I made that I don't think we would have been able to get, you know, the legal and privacy appetite to handle risk when we already had such a huge backlog of cases to review. 9 Ω. Okay. So the previous document we 10 looked at indicated there were about 450,000 11 reports. Here it says over 2 million reports. 12 Do you have an understanding of why 13 there was that growth, or are these just measuring 14 different things? 15 MR. SNEED: Object to form. 16 Characterization. 17 18 THE WITNESS: I don't have a reference as to why there was that growth, but it 19 also wouldn't surprise me. We have, you know, so 20 21 many users on Instagram and Facebook. doesn't necessarily seem like a huge jump in that It that 22 period of time. 23 BY MR. WARREN: 24 Ω. Okay. Do you know if the 25 2.5 million was referring to the whole family of 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 50 of 211 Page 98 1 apps monitored by Meta as opposed to just 2 Instagram? 3 MR. SNEED: Object. Foundation. 4 THE WITNESS: I don't know which 5 specific subset it was referring to. 6 Given that primarily worked 7 8 on Facebook, it might have just been Facebook. it could have been across both Facebook and But 9 Instagram. 10 BY MR. WARREN: 11 Ω. Okay. Regardless, your 12 13 14 15 understanding is, as of this point in time, there were somewhere between half a million to 2.5 million reports of under-13 users on META Platforms, right? 16 MR. SNEED: Object to form. 17 Characterization. 18 THE WITNESS: Yes. 19 BY MR. WARREN: 20 Ω. 21 Okay. And that was a backlog that Meta had not yet gotten around to reviewing, right? 22 A. Yes. 23 MR. SNEED: Same objections. 24 BY MR. WARREN: 25 Ω. 's description of a 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 51 of 211 Page 99 1 seven-screen-long reporting option seems to me to 2 be pretty clunky. 3 4 How does it seem to you? MR. SNEED: Object to form. THE WITNESS: That is really 5 6 clunky. Generally for seamlessness of the user experience, we don't typically recommend more than 7 two or three screens. The more screens you have, 8 the greater the drop-off of reports, because it's 9 10 just too much work for people to keep going through the process to report something. 11 BY MR. WARREN: 12 Ω. Got it. 13 Okay. So I want to just put these 14 pieces together before we move on. 15 Until December 2019, Meta didn't 16 17 even ask kids on Instagram for their birthdays, right? 18 MR. SNEED: Object to form. 19 THE WITNESS: Yes. 20 BY MR. WARREN: 21 Ω. And when Meta finally started 22 asking that question, it learned that lots of kids 23 lie, right? 24 MR. SNEED: Same objection. And 25 foundation. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 52 of 211 Page 100 1 THE WITNESS: Yes. 2 BY MR. WARREN: 3 Ω. So as a belt-and-suspenders 4 5 measure, Meta gave people the option to report under-13 users, but it didn't do that on the 6 Instagram app itself, and it made the reporting 7 option seven screens long, right? 8 MR. SNEED: Object. Compound. And 9 foundation. 10 THE WITNESS: I don't know if 11 the - I don't know when Instagram started offering 12 13 14 the option to report under-13 users. It could have been before they started collecting age. I don't know. 15 BY MR. WARREN: 16 Ω. Okay. In any event, the reporting 17 option that existed as of March 2021 was 18 insufficiently clean and clear to allow people to 19 easily report underage users, right? 20 MR. SNEED: Object to the 21 characterization. Foundation. 22 THE WITNESS: In my opinion, I 23 think it was a really complicated process. 24 And just, you know, as someone who 25 has been in the industry for a long time, it didn't 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 53 of 211 Page 101 1 2 3 seem to be at par with what, you know, other industry players would typically offer as a reporting option. 4 BY MR. WARREN: 5 Ω. 6 Sure. And if Meta had committed more 7 8 resources to the project, could it have cleared out that backlog? 9 MR. SNEED: Object to form. Calls 10 for speculation. 11 THE WITNESS: I think - I think we 12 could have reviewed those cases with more 13 resourcing. 14 BY MR. WARREN: 15 Ω. Okay. All right. I'm going to 16 17 introduce an exhibit to address one of counsel's foundation objections really quickly. And this 18 will be kind of a standalone. 19 So I'm going to introduce an 20 exhibit that we'll mark as Exhibit 9 and is it 21 Exhibit 10? This will be Exhibit 10, I'm sorry. 22 MR. DRAKE: You marked Exhibit 8 23 for the record. So that got marked and then you 24 went to 9. So I marked it 8, 9. If it's okay, we 25 can use that numbering and then clear it up at a 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 54 of 211 Page 132 1 2 3 be marked as Exhibit 14. This is a document bearing the Bates stamp META3047MDL-040-00541333. (Document marked for identification 4 as Jayakumar Exhibit 14.) 5 BY MR. WARREN: 6 Ω. Okay. This appears to be a chat 7 between you and ? 8 A. Mm-hmm, yes. 9 Ω. And it looks like the chat is from 10 November 23, 2021; is that right? 11 A. Yes. 12 Ω. Who is Ms. ? 13 A. was a researcher at Meta. 14 Ω. Okay. And do you know what type of 15 research she worked on in particular? 16 A. She worked on a number of the 17 safety and well-being research products. This 18 conversation was in relation to a talk that she was 19 giving where she talked about how she used to work 20 on sleep products at Bose. 21 Ω. Okay. Did you participate in this 22 chat during the ordinary course of your work at 23 Meta? 24 A. Yes. 25 Ω. Okay. Could you read your opening 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 55 of 211 Page 133 1 message to please? 2 A. "I didn't know you used to work on 3 4 more broadly. 5 sleep products. We should chat about well-being There are a number of sleep-promoting interventions I want to urge Meta 6 towards." 7 Ω. Why did you send that note to 8 ? 9 A. She had just spoken about how she 10 11 12 used to work on sleep products, and some of their research around better sleep that could potentially be translated into what we do. 13 And I had responded to her saying 14 15 that, you know, from a practitioner perspective, as a person kind of developing strategies and ideas 16 17 18 19 for product and policy teams, I really wanted to pitch a number of sleep-promoting interventions to improve the well-being of young people on the platform. 20 Ω. Okay. And responds that 21 22 23 24 A. 25 Ω. she worked for about four years at Bose on sleep products. She says people have awful sleep habits. 877-370-3377 Do you see all that? Yes. And what did you say in response? Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 56 of 211 Page 134 1 A. I said that: 2 3 4 5 6 I would love to brainstorm on what we can do across the family of apps about this. Social media addiction is such a problem, and I optimistically think we have a lot of low-hanging fruit we can pick here. 7 Ω. Okay. Now, this chat was about 8 nine months after had told you 9 10 that in her view social media addiction was not a companywide concern for Meta, correct? 11 A. Correct. 12 MR. SNEED: Object to form. 13 BY MR. WARREN: 14 Ω. So none of that low-hanging fruit 15 had been picked earlier in 2021 by Meta, right? 16 MR. SNEED: Object to form. 17 THE WITNESS: Yeah. None of them 18 had been picked. 19 BY MR. WARREN: 20 Ω. Okay. Ms. Jayakumar, if you knew 21 22 23 24 25 that the risk of social media addiction had been raised to Meta executives in 2016, would it surprise you that in 2021, there was still low-hanging fruit to address that issue? 877-370-3377 MR. SNEED: Object to form and Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 57 of 211 Page 135 1 foundation. 2 THE WITNESS: Would it surprise me, 3 no. 4 BY MR. WARREN: 5 Ω. Why wouldn't it surprise you? 6 A. I could imagine I mean I 7 8 9 10 11 12 13 14 15 16 17 Ω. 18 19 20 21 there are many things that are raised through senior leadership, you know, over the course of, say, five years. And I think it would really depend on how much external pressure there was. How much regulatory pressure there was to do something in the space. How much advertising business pressure there was. And if those pressures didn't exist, then I wouldn't be surprised if the low-hanging fruit hadn't been picked in 2021. If you knew the risk of social media addiction had been raised to Meta executives in 2016, do you think Meta should have picked the low-hanging fruit to address that problem? MR. SNEED: Object to form and 22 foundation. 23 THE WITNESS: Yes, I do. 24 Yes, I do. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 58 of 211 Page 136 1 Ω. At any point while you worked at 2 Meta, did the company warn its user base that 3 Instagram could be addictive? 4 A. No. 5 Ω. At any point while you worked at 6 7 Meta, did it warn its user base of the risk of problematic use of Instagram other than addiction? 8 MR. SNEED: Object to form. 9 THE WITNESS: No. And I don't know 10 that there was an acceptance, you know 11 12 13 externally, that there was any acceptance that there were ways in which an app could be used problematically. 14 BY MR. WARREN: 15 Ω. But internally that was a 16 conversation that researchers were having, correct? 17 MR. SNEED: Object to form and 18 foundation. 19 20 THE WITNESS: Yes. It's a conversation internally researchers were having. 21 But I don't know if I don't think it's something 22 that there was consensus on within the broader Meta 23 leadership. 24 So researchers, data scientists, 25 product policy teams, we can do a lot of work and 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 59 of 211 Page 137 1 2 3 we can bring a lot of evidence to the table that is not equivalent to having volume or support for that hypothesis at a leadership level. 4 BY MR. WARREN: 5 Ω. Okay. So because Meta leadership 6 had not bought into the data and evidence 7 concerning problematic use, they did not warn 8 Instagram's user base about that issue, correct? 9 MR. SNEED: Object to form. 10 Characterization. 11 THE WITNESS: That would be -- that 12 would be my impression. 13 BY MR. WARREN: 14 Ω. Okay. So let me ask you about the 15 low-hanging fruit that you had in mind. And let's 16 start with the recommendation algorithm. 17 In your experience, did the 18 algorithm, as deployed in the Explore and Reels 19 parts of Instagram, contribute to social media 20 addiction? 21 MR. SNEED: Object to form and 22 foundation. 23 24 THE WITNESS: I think it played one role. I think social media addiction is a really 25 complex issue. And I don't want to characterize 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 60 of 211 Page 138 1 any one app as solely driving it. But yes, I think 2 it played a role. 3 BY MR. WARREN: 4 Ω. And what was the role that it 5 played? 6 A. I think the design of the algorithm 7 to promote sustained engagement by continuously 8 showing appealing content and increasingly 9 appealing content can play a role in drawing 10 11 viewers into the experience for longer periods of time than they necessarily would want for 12 themselves. 13 That said, there are other factors, 14 of course, that can play a role. 15 know, as somebody who has been in the industry for And I want you 16 17 18 a long time, I want to be really clear that it's not a single deciding factor but it does have a role to play, yes. 19 Ω. Okay. And in your experience, did 20 21 22 the recommendation algorithm, as deployed in the Explore and Reels part of Instagram, contribute to other kinds of problematic use besides addiction? 23 MR. SNEED: Object to form and 24 foundation. 25 877-370-3377 THE WITNESS: I think that there Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 61 of 211 Page 157 1 as Jayakumar Exhibit 16.) 2 3 MR. WARREN: This is a document bearing the Bates stamp META3047MDL-040-00337135. 4 BY MR. WARREN: 5 Ω. 6 And I'll represent to you that this document was produced by Meta and it came from your 7 electronic files. Do you have any reason to 8 dispute that? 9 A. No. 10 Ω. What's the title of this document? 11 A. "AR Beauty Effects: Stakeholder 12 Engagement." 13 Ω. Was this document created by 14 employees of Meta? 15 A. Yes. 16 Ω. Did you review this document while 17 you were working at Meta? 18 A. Yes. 19 Ω. Okay. And can you remind the jury 20 what AR stands for? 21 A. Augmented reality. 22 Ω. Okay. Please direct your attention 23 to the bottom of the first page where it says, 24 "Feedback Summary." 25 Do you see that? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 62 of 211 Page 158 1 A. Yes. 2 Ω. 3 Does this section of the document summarize the feedback that Meta received from 4 5 experts concerning augmented reality beauty filters? 6 MR. SNEED: Object to form and 7 foundation. 8 THE WITNESS: Yes, it does. 9 BY MR. WARREN: 10 Ω. Okay. Can you please read the 11 first two bullets? 12 A. "Generally, stakeholders support 13 14 15 making a distinction between effects achieved via makeup versus those that can only be achieved via cosmetic surgery, though no line will be perfect." 16 Ω. And what about the next bullet? 17 A. "These extreme beauty effects can 18 have severe impacts on both the individuals using 19 the effects and those viewing the images." 20 2. Do you agree with that statement? 21 A. Yes, I do. 22 Ω. Can you read the last bullet, the 23 one beginning with "children"? 24 A. "Children are particularly 25 vulnerable, however many others are vulnerable as 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 63 of 211 Page 159 1 2 well: those with a history of mental health challenges, eating disorders, et cetera." 3 Ω. Okay. Do you agree with that? 4 A. Yes. 5 Ω. All right. Can you please flip the 6 page. Do you see the heading titled, "Summary for 7 Boz e-mail"? 8 A. Yes. 9 Ω. Who is Boz? 10 A. This is Andrew Bosworth. At the 11 time he was the head of the VR team. 12 Ω. Was he a senior leader at Meta? 13 A. Yes. Very senior. 14 MR. SNEED: Object to form. 15 BY MR. WARREN: 16 Ω. Can you please read the first 17 paragraph of this summary? 18 A. 19 20 21 22 23 24 25 "We spoke with 18 experts including psychologists, researchers, body image activists, and AR professionals. We made sure to focus our outreach on experts who could provide research insights and findings about the impact of social media and internet technologies on youth identity development, body image, including Body Dysmorphic Disorder (colloquially know as 'Snapchat 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 64 of 211 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 Page 160 dysmorphia'), and mental health. A large majority, including individuals in the AR field, recommended prohibiting these filters, citing known impacts to body image and mental health of other forms of media that idealize unrealistic beauty standards. We discussed corrective actions that were less prohibitive, including limiting cosmetic surgery filters to adults and/or utilizing watermarks, however these actions were not viewed as sufficient protections because (1) the universe of users who are vulnerable to negative impacts from such filters is much broader than those ages 13 to 17, and (2) past research on labeling photoshopped imagery found that they do not insulate viewers from its negative affects. There was agreement in our conversations that we should be careful to limit cosmetic filters even when used in conjunction with a costume, animal features, fantasy effects and accessories, especially because these are often most popular with our youngest users who are attracted to the fun/imaginary play aspect of these filters but are gradually exposed and increasingly grow accustomed to their idealized face over their real one." 877-370-3377 Ω. Do you take issue with anything in Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 65 of 211 Page 161 1 that paragraph? 2 MR. SNEED: Object to form. 3 BY MR. WARREN: 4 Ω. Well, let me ask it a different 5 way. I'll withdraw the question. 6 Is it your understanding that these 7 8 findings were put into an e-mail that was sent to Andrew Bosworth? 9 MR. SNEED: Object to form and 10 foundation. 11 THE WITNESS: That's my 12 understanding, yes. 13 This predates my time at the 14 company though, but it is my understanding. 15 BY MR. WARREN: 16 Ω. Did Meta leadership follow the 17 feedback it received from the 18 experts that it 18 consulted? 19 MR. SNEED: Object to form and 20 foundation. 21 THE WITNESS: No, it did not. 22 BY MR. WARREN: 23 Ω. Who made that decision? 24 MR. SNEED: Object to form and 25 foundation. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 66 of 211 Page 162 1 THE WITNESS: I believe Boz made 2 the final decision. 3 BY MR. WARREN: 4 Ω. What was your involvement with 5 respect to this issue? 6 A. I joined after the consultations 7 that happened with this list of experts in 2019. 8 And I was there when an escalation 9 around this topic took place. 10 I wasn't closely involved in it 11 12 because it was pretty early in my time at the company. 13 And then I received the direction 14 15 that Boz had overturned the initial decision to move forward with removing these filters from the 16 store. 17 My role after that was then to go 18 back 19 20 was to actually develop a set of arguments around why it might be a good thing to keep these filters in store. 21 However, while keeping them in 22 23 store, we would not recommend them actively. So they would just be available but not recommended. 24 And then my role was to develop 25 arguments as to why this is a better option than 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 67 of 211 Page 163 1 2 3 4 removing them altogether, and to go back to these experts and talk to them about our decision and make sure that they were kept updated about the direction we had gone in. 5 Ω. 6 7 Okay. So is it fair to say that you were asked to help effectuate Andrew Bosworth's decision to keep cosmetic beauty filters on 8 Instagram? 9 MR. SNEED: Object to form. 10 Characterization. 11 12 THE WITNESS: I was with socializing the decision with our community of I was tasked 13 experts. 14 BY MR. WARREN: 15 Ω. Did you agree with the decision? 16 A. I did not. 17 Ω. Did Meta leadership make the right 18 decision in restoring augmented reality cosmetic 19 beauty filters to Instagram? 20 MR. SNEED: Object to form and 21 foundation and vague. 22 THE WITNESS: I do not think that 23 Meta made the right decision there, no. 24 BY MR. WARREN: 25 Ω. And why is that? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 68 of 211 Page 164 1 A. I'm sorry? 2 Ω. Why is that? 3 MR. SNEED: Object to form. 4 THE WITNESS: I think we heard a 5 6 7 good amount of feedback both from experts, as well as a good amount of research and community feedback around how people were feeling about beauty 8 filters. 9 Purely from - I mean, my expertise 10 is in safety and well-being. So from a safety and 11 well-being perspective, I didn't think that this 12 was the right approach to take. 13 I could understand the other 14 15 16 reasons, and you know, the other kind of motivations around it. But from a safety and well-being perspective, I didn't think that this 17 was the right decision. 18 BY MR. WARREN: 19 Ω. Do you have an understanding as to 20 why this decision was made? 21 MR. SNEED: Object to form and 22 foundation. 23 THE WITNESS: My understanding, and 24 25 the this is really from what I was told, you know, from about Boz's decision, which was indirectly 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 69 of 211 Page 165 1 communicated. 2 But my understanding is that AR 3 these AR filters are incredibly popular, and we do 4 see a lot of use about of these filters. They 5 are also really popular on other platforms like 6 Snap and TikTok and so that actually Snap more 7 than TikTok. 8 And I think banning them would have 9 felt like a competitive loss in contrast to, you 10 know, someone who was actively using these features 11 and very successfully using these features. 12 BY MR. WARREN: 13 Ω. So a few minutes ago I previewed 14 15 16 that I wanted to discuss with you three different features related to negative social comparison. talked about the first one, beauty filters. We 17 A. Mm-hmm. 18 Ω. And do you agree that well, let 19 me withdraw the question. 20 I'd like to ask you about the 21 second one, the public display or tally of likes. 22 So to start, can you explain to the 23 jury what a like is? 24 A. A like is a reaction that someone 25 can make on the app to a piece of content that 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 70 of 211 Page 166 1 suggests that they saw, acknowledged, or liked this 2 content in some way. 3 Ω. Okay. So let's say I post a video 4 on Reels. Will I be able to see how many people 5 liked that video? 6 A. Yes, you will. 7 Ω. Hopefully lots. 8 Will other users be able to see how 9 many likes my video received? 10 A. Yes. 11 Ω. At some point did researchers at 12 Meta determine that that feature may be 13 exacerbating negative social comparison for kids? 14 MR. SNEED: Object to form and 15 foundation and characterization. 16 THE WITNESS: Yes. Researchers 17 18 have put together a really compelling package of information based on their interviews and user 19 research suggesting that I'm sorry suggesting 20 21 that likes might be one way in which we could target the issue of negative social comparison. 22 They had identified a number of 23 other methods as well, and likes was one of them. 24 BY MR. WARREN: 25 Ω. Who is Adam Mosseri? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 71 of 211 Page 167 1 A. He is the head of Instagram. 2 Ω. Also a fairly senior leader at 3 Meta? 4 A. Yes. 5 Ω. Did Adam Mosseri announce at some 6 7 8 point that Instagram would be launching a new safety feature to address the problem of visible likes exacerbating negative social comparison? 9 MR. SNEED: Object to form. 10 THE WITNESS: Yes. He announced it 11 in 2019. 12 BY MR. WARREN: 13 Ω. 14 A. 15 Ω. 16 And what was the project called? It was called Project Daisy. What happened to Project Daisy? MR. SNEED: Object to form. Vague. 17 Foundation. 18 THE WITNESS: Project Daisy 19 20 21 22 eventually launched in a very different form than what had originally tested. What had originally been tested was turning off the like count in general for all users. 23 What eventually launched was 24 25 including the option to turn off likes in the settings tab and letting users go in and decide 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 72 of 211 Page 168 1 whether they wanted to hide likes or not. 2 BY MR. WARREN: 3 Ω. Was the version of Project Daisy 4 5 that launched as protective of youth safety as the version that was under consideration? 6 MR. SNEED: Object to form and 7 foundation and calls for speculation. 8 THE WITNESS: I think that the 9 10 11 version of Project Daisy that launched did not address the core issue of negative social comparison that the original version of the product 12 was trying to target. 13 BY MR. WARREN: 14 Ω. And why do you think that? 15 A. We've previously talked about how 16 17 18 19 20 21 most people don't go into their settings and change them unless really they are, you know, in a moment of crisis or trying to protect their safety or well-being in some active way, because they've encountered some harm. In this case, because the option 22 was simply in the settings, we didn't anticipate, 23 and I don't think nobody people didn't really 24 25 And so the issue of go in and adjust those settings in any significant quantities. the goal of Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 73 of 211 Page 178 1 that's been marked for identification as 2 3 Ms. Exhibit 17. This appears to be a chat between you from April 9, 2021. 4 Do you agree with that? 5 A. Yes. 6 Ω. Were these chats sent in the 7 ordinary course of your work at Meta? 8 A. Yes. 9 Ω. Who is ? 10 A. at the time was on the 11 Instagram public policy team. 12 Ω. In this conversation, were you and 13 Ms. discussing how Instagram's recommendation 14 of content could exacerbate eating disorders? 15 A. Yes. 16 Ω. At 10:03 a.m., Ms. asks you: 17 Would we consider not surfacing 18 certain content in our rec surfaces that we know to 19 be triggering? 20 Do you see that? 21 A. Yes. 22 Ω. What are rec surfaces? 23 A. These are places where Instagram 24 25 Ω. recommends content, like the Explore tab or Reels. 877-370-3377 And you respond: Golkow Technologies, A Veritext Division www.veritext.com 1 2 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 74 of 211 Page 179 Totally we're working on a borderline ED policy right now to address this and 3 are just kicking this off. 4 Did I read that correctly? 5 A. Yes. 6 Ω. Ms. then says at 10:06 a.m.: 7 The image that sent 8 yesterday of an Explore feed is really problematic, 9 and the ownership of that is also on us since it's 10 an unconnected recommended surface. 11 Did I read that correctly? 12 A. Yes. 13 Ω. And we talked about this briefly, 14 but how did you understand Ms. when she 15 referred to an unconnected recommended surface? 16 MR. SNEED: Object to form. 17 18 19 20 THE WITNESS: Recommendation surfaces include content from people whom you may not follow or accounts that you may not know. And that's what we mean by an unconnected surface. 21 BY MR. WARREN: 22 Ω. Do you recall the images that 23 Ms. was referencing here? 24 A. I don't recall the specific images 25 that sent us, no. Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 75 of 211 Page 180 1 Ω. Okay. So I'm going to show you 2 those images. And I've actually handed you a 3 document bearing the Bates stamp META3047MDL-145- 4 00000001. 5 And that's been marked for 6 identification as Exhibit 18. 7 (Document marked for identification 8 as Jayakumar Exhibit 18.) 9 BY MR. WARREN: 10 Ω. 11 A. Do you have that in front of you? I do. 12 Ω. And I'll represent to you that 13 there are some files that are shared in the course 14 15 of this chat, and those are attached to the back of this document. 16 A. Okay. 17 Ω. Any reason to dispute that? 18 A. No. No reason. 19 Ω. Okay. And you can see about 20 halfway down sorry, withdrawn. 21 22 This appears to be a chat thread from April 8, 2021, between you, Ms. and 23 ; is that correct? 24 A. Yes. 25 Ω. 877-370-3377 Okay. And you can see about Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 76 of 211 Page 181 1 halfway down, Mr. says: 2 The question is that there is 3 4 content that will continue to be created on the platform that makes teens feel back about 5 themselves. 6 I assume he meant bad, and that was 7 a typo. Do you agree? 8 A. Yes. 9 Ω. Okay. And did I read that 10 correctly? 11 A. Yes. 12 Ω. And then he shares some 13 screenshots. And he asks, "What can we do about 14 that"; is that right? 15 A. Yes. 16 Ω. Okay. And can you flip to the 17 attachments? 18 They are up on the screen as well. TRIAL TECH: Do you want them up 19 side by side, two pages? 20 MR. WARREN: Yeah. 21 BY MR. WARREN: 22 Ω. How would you describe the Explore 23 feeds that are displayed here? 24 A. They are predominately feminine, 25 mostly women mostly young women or young girls. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 77 of 211 Page 182 1 2 3 4 5 6 7 you know They are indexed largely towards, well, there are a number of pieces of content with young girls or women in minimal clothing. A lot of them are kind of self-image or representations, either selfies or photos taken of themselves in mirrors. And they are predominately white 8 images. There are very few people of color in 9 these images. 10 Ω. And what about the Explore screen 11 on the right? 12 A. Explore screen on the right. So... 13 MR. SNEED: Are you referring to 14 ending in Page 5? 15 THE WITNESS: These two are the 16 same. 17 BY MR. WARREN: 18 2. There should be another page. 19 TRIAL TECH: I think everybody got 20 the same. 21 THE WITNESS: This - this screen 22 23 24 25 shot has a number of images featuring cosmetic procedures or makeup. procedures. BY MR. WARREN: Yeah, a number of cosmetic 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 78 of 211 Page 183 1 Ω. Okay. So let's go back to the 2 prior exhibit if we can, that's Exhibit 17. 3 A. Mm-hmm. 4 Ω. And this is your conversation with 5 Ms. So, again, at 10:06 a.m., Ms. 6 says: 7 8 9 The images that yesterday of an Explore feed is really problematic. And are the images we just looked sent 10 at what Ms. was referencing? 11 MR. SNEED: Object to form. 12 THE WITNESS: The images she's 13 referencing. Yes, I think so. 14 BY MR. WARREN: 15 Ω. Okay. Do you agree those are 16 really problematic? 17 A. I think they could be problematic. 18 I wouldn't say definitively that they are 19 problematic. 20 Ω. And under what circumstances would 21 they be problematic? 22 A. I think social comparison in 23 24 25 particular, and, you know, eating disorders is a really complex space. So it really would depend on a number of factors, including the preexisting 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 79 of 211 Page 184 1 2 3 4 5 conditions of the people who were viewing that content. One could argue that if you already struggle with negative social comparison, being flooded with a wave of images like that could be could make you feel really bad about yourself. 6 But it's not a given that anyone 7 8 9 who looks at those images would necessarily feel bad about themselves or compare themselves to one another in a negative way. 10 It also depends on how often that 11 is the composition of their Explore tab. 12 So if you log into Instagram once 13 14 15 16 17 18 and that's your Explore tab, and then, you know, you log in another three times and your Explore tab looks different, I don't think that would have as much of an impact on social comparison as if, say, every time you opened your Instagram app, you were getting flooded with aggregate set of images. 19 Ω. And you've worked on youth safety 20 for over a decade, correct? 21 A. Yes. 22 Ω. In your experience, do teenage 23 24 girls have a higher degree of social comparison than the median population? 25 MR. SNEED: Object to form and Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 80 of 211 Page 185 1 foundation. 2 THE WITNESS: Yes. Teenage girls 3 generally -- 4 well, teenage girls report having higher levels of negative social comparison. That 5 6 doesn't necessarily mean that they have more than, say, teenage boys. 7 But certainly more than the median, 8 say, adult population. 9 BY MR. WARREN: 10 Ω. Okay. So the Explore feeds we just 11 looked at are more likely to be problematic for 12 13 teenage girls than, say, adult men, would you agree? 14 MR. SNEED: Object to form and 15 foundation and vague. 16 THE WITNESS: They might have a 17 higher likelihood of that, yes. 18 BY MR. WARREN: 19 Ω. Okay. And if the Explore feed of a 20 21 22 23 teenage girl is composed of pages that look like that, then it's more likely that could create problems of negative social comparison, would you agree? 24 MR. SNEED: Object to form and 25 foundation. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 81 of 211 Page 186 THE WITNESS: More likely than 1 2 what? 3 BY MR. WARREN: 4 Ω. More likely than if that was just a 5 one off. 6 MR. SNEED: Object to form and 7 foundation. 8 THE WITNESS: Yes. 9 BY MR. WARREN: 10 Ω. Okay. So Ms. says at the end 11 of that sentence: 12 The ownership of that is also on 13 us. 14 What did you understand her to mean 15 when she said that? 16 MR. SNEED: Object to form and 17 foundation. 18 THE WITNESS: 19 meant by that, and I I think what she this is just my opinion 20 21 22 23 24 25 based on what she's written, her point was that these are not accounts that people have chosen to follow or that even, you know, young people have chosen to follow. If they had chosen to follow them, you could argue that, you know, they made that choice to see that content. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 82 of 211 Page 187 But in these cases, we are 1 2 3 4 5 recommending this content that's from people that you don't follow, and, therefore, the responsibility of making sure those are responsible recommendations falls to us. 6 BY MR. WARREN: 7 Ω. And did you agree with Ms. ? 8 A. Yes. 9 Ω. Now, you responded to Ms. by 10 saying: 11 Yeah the challenge is that when 12 you zoom in, any one of those photos probably 13 wouldn't be triggering, but in an Explore grid with 14 15 all these photos stacked against one another, it's pretty overwhelming. 16 Did I read that correctly? 17 A. Yes. 18 2. And can you explain that? 19 A. The problem with the difficulty 20 21 with moderating aggregated content or surfaces with aggregated content is that any one of those pieces 22 23 24 of content is not going to violate any policy. It's not even necessarily going to be borderline violative of any policies. 25 But when you combine them in Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 83 of 211 Page 188 1 2 aggregate, and, you know, you've got dozens of these images in one page on the Explore tab, that 3 gets - that can be a really overwhelming 4 experience. 5 And at the time, this was in 2021, 6 7 8 9 10 I don't think we had done enough work to, like, understand or mitigate the impact of aggregated surfaces. And there were a number of ways in which we hadn't done enough work on them. And I think the Explore grid was a really good example of that. 11 Ω. Okay. And during the three and a 12 13 14 half years that you worked at Meta, did it devote the technical resources and the time necessary to address this problem? 15 MR. SNEED: Object to form. 16 THE WITNESS: I think there was 17 some efforts to address these problems. For 18 example, the borderline ED policy that I mention in 19 here. 20 And we had very supportive product 21 partners. But due to just overall under-resourcing 22 23 of this area, we certainly couldn't move as fast or as extensively as we would have liked, given the 24 solutions that were available to us. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 84 of 211 Page 189 1 Ω. And let me just explore what you 2 just said. 3 So you were referring to a 4 borderline policy; is that right? 5 A. Mm-hmm, yes. 6 Ω. And that refers to content that may 7 not violate the letter of Instagram's community 8 standards, but it may still be something that you 9 want to demote or remove; is that right? 10 A. Yes. 11 MR. SNEED: Object to form. 12 BY MR. WARREN: 13 Ω. Okay. But a borderline policy 14 would not address problems arising from the 15 16 aggregation of content that isn't even borderline, right? 17 MR. SNEED: Object to form. 18 THE WITNESS: A borderline policy 19 would not address the problems that arise from 20 aggregated content that isn't borderline. 21 22 However, if you expand the borderline policy to include more types of content, 23 then you could make an impact on recommendation 24 surfaces. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 85 of 211 Page 190 1 Ω. Okay. Let's take dieting tips as 2 an example. 3 A. Right. 4 Ω. Okay. Dieting tips do not violate 5 Instagram's community standards, correct? 6 A. They do not. 7 Ω. And they don't violate Instagram's 8 borderline policy correct? 9 A. The current they don't violate 10 the current iteration of the borderline policy, 11 yes. But we could we could change that. 12 Ω. Right. But when you were working 13 14 there, they did not violate the borderline policy, correct? 15 A. Right. But that was the work we 16 17 18 19 were doing to try and see if we could expand the borderline policy to include more types of content that we would say, this is content that we just don't want to recommend. 20 Ω. Right. And the policy was never 21 actually expanded to cover dieting tips, correct? 22 MR. SNEED: Object to form and 23 foundation. 24 THE WITNESS: Yes. That's correct. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 86 of 211 Page 191 Okay. So if an Explore grid for a 1 Ω. 2 3 4 5 teenage girl consists entirely of dieting tips, that would not violate Instagram's community standards and it would not violate Instagram's borderline policies, correct? 6 MR. SNEED: Object to form, vague. 7 THE WITNESS: Correct. 8 BY MR. WARREN: 9 Ω. But that could still create a 10 11 problem for that individual because of the aggregated nature of that content, correct? 12 MR. SNEED: Object to form. 13 THE WITNESS: Yes, definitely. 14 BY MR. WARREN: 15 Ω. And that's not a problem that 16 Instagram devoted the technical resources to fix 17 while you were there, correct? 18 MR. SNEED: Object to form and 19 foundation. 20 THE WITNESS: That's correct. 21 BY MR. WARREN: 22 Ω. Okay. So let's continue on with 23 the chat with you and Ms. 24 At 10:13 a.m. she says: 25 Girls as young as 8 start to lose Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 87 of 211 Page 192 1 2 self-confidence because of negative self-image and social comparison. It's when they become aware of 3 their bodies versus others. 4 Did I read that correctly? 5 A. Yes. 6 Ω. Do you agree with what Ms. 7 wrote? 8 A. Yes, I do. 9 Ω. Can you read your response please? 10 A. I said, "Yeah and we're playing the 11 same role today in their perception of 12 self-identity that magazines and music videos 13 played when we were growing up." 14 Ω. Do you stand by that? 15 A. I do. 16 Ω. Did Meta ever warn kids that 17 Instagram uses their data to aggregate and push 18 content that can harm how they see themselves? 19 MR. SNEED: Object to form. 20 Characterization. And foundation. 21 THE WITNESS: No, I don't think so. 22 BY MR. WARREN: 23 Ω. I want to direct your attention to 24 the second page of this document. 25 At 10:18 a.m. Ms. says: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 88 of 211 1 2 3 4 5 Page 193 There are really meaty legal issues separate and apart of the policy and public health concerns. Largely by way of deceptiveness (which I think is something that we should start leveraging more). 6 Did I read that correctly? 7 A. Yes. 8 Ω. Is it correct that Ms. was 9 10 11 suggesting to you that if Instagram didn't address this problem, it could create legal issues related to deceptiveness? 12 MR. SNEED: Object to form, 13 foundation, and calls for a legal conclusion. 14 THE WITNESS: This was nearly four 15 years ago. But I believe that is accurate. 16 BY MR. WARREN: 17 Ω. Okay. So before we move on and 18 we'll take a quick break I'd like your help 19 summarizing where we are. 20 So you've now testified, I believe, 21 22 23 that Instagram was aware that some of its younger users could experience negative social comparison through use of the platform, correct? 24 25 MR. SNEED: Object to form. THE WITNESS: Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 89 of 211 Page 194 1 BY MR. WARREN: 2 Ω. Meta heard from experts that 3 4 cosmetic beauty filters contributed to that problem, correct? 5 MR. SNEED: Object to form. 6 THE WITNESS: Yes. 7 BY MR. WARREN: 8 Ω. But instead of banning those 9 filters, it allowed them to continue being used on 10 Instagram, correct? 11 MR. SNEED: Object to form. 12 THE WITNESS: Yes. 13 BY MR. WARREN: 14 Ω. Meta also heard from internal 15 researchers that publicly visible like counts 16 contributed to negative social comparison, correct? 17 18 MR. SNEED: Object to form. THE WITNESS: They heard that 19 negative 20 21 publicly visible like counts could contribute and could be a factor in contributing to negative social comparison. 22 BY MR. WARREN: 23 Ω. And Instagram explored changing 24 that feature in a way that would have mitigated the 25 problem, correct? Golkow Technologies, 877-370-3377 A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 90 of 211 Page 195 1 MR. SNEED: Object to form. 2 THE WITNESS: Yes. 3 BY MR. WARREN: 4 Ω. But it didn't actually roll out the 5 6 version of that feature that would have helped; instead it did something watered down, correct? 7 MR. SNEED: Object to form and 8 characterization. 9 THE WITNESS: Yes, correct. 10 BY MR. WARREN: 11 Ω. And you also testified about how 12 Instagram's recommendation algorithm can contribute 13 to negative social comparison and actually 14 exacerbate eating disorders. 15 Do you agree with that? 16 MR. SNEED: Object to form. 17 THE WITNESS: Yes. 18 BY MR. WARREN: 19 Ω. And you testified, during your 20 three and a half years at the company, you flagged 21 how that was an issue on Explore and Reels, 22 correct? 23 MR. SNEED: Object to form. 24 THE WITNESS: Yes, I did. 25 BY MR. WARREN: 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 91 of 211 Page 196 1 Ω. But Meta did not invest sufficient 2 resources to solve that problem either, correct? 3 MR. SNEED: Object to form. 4 THE WITNESS: Yes. 5 MR. WARREN: Okay. I'm going to 6 move into another unit. Why don't we take a break. 7 MR. SNEED: Okay. 8 THE VIDEOGRAPHER: The time right 9 now is 2:10 p.m. We are off the record. 10 (Short break.) 11 THE VIDEOGRAPHER: The time right 12 now is 2:21 p.m. We're back on the record. 13 BY MR. WARREN: 14 Ω. Okay. Ms. Jayakumar, I'd like to 15 ask you some questions now about a difficult topic. 16 It maybe difficult to discuss and difficult for the 17 jury to hear about, but it's an important one. 18 Are you familiar with the acronym 19 SSI? 20 A. Yes. 21 Ω. And what does that stand for? 22 A. It stands for suicide and 23 self-injury. 24 25 (Document marked for identification as Jayakumar Exhibit 19.) 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 92 of 211 Page 217 1 THE WITNESS: Yes. 2 BY MR. WARREN: 3 Ω. Okay. I've handed you Exhibit 22 4 and Exhibit 23. Are you familiar with these 5 documents? 6 A. Yes. 7 Ω. Okay. Is Exhibit 22 a copy of 8 Meta's policy on suicide, self-injury, and eating 9 disorders from May 4, 2020? 10 A. Yes. 11 Ω. And is Exhibit 23 a copy of Meta's 12 policy on suicide, self-injury, and eating 13 disorders from January 2025? 14 MR. SNEED: Object to form. 15 THE WITNESS: Yes. 16 BY MR. WARREN: 17 Ω. Okay. And I'd ask you to just 18 19 compare the first three paragraphs in the 2020 and 2025 versions of the policy, and let the jury know 20 if they are the same. 21 A. Yes. They are the same. 22 Ω. 23 24 So this aspect of Meta's suicide, self-injury, and eating disorders policy hasn't changed since roughly around the time you started 25 working at Meta, correct? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 93 of 211 Page 218 1 MR. SNEED: Object to form. 2 Characterization. 3 THE WITNESS: Yes. 4 BY MR. WARREN: 5 Ω. Okay. Can you please read the 6 second paragraph of the policy to the jury? 7 A. 8 9 10 11 12 13 "While we do not allow people to intentionally or unintentionally celebrate or promote suicide, self-injury or eating disorders, we do allow people to discuss these topics because we want our services to be a space where people can share their experiences, raise awareness about these issues, and seek support from one another." 14 Ω. Please read the first sentence of 15 the next paragraph. 16 A. "We remove any content that 17 encourages suicide, self-injury or eating 18 disorders, including fictional content such as 19 memes or illustrations, and any self-injury content 20 which is graphic regardless of context." 21 Ω. Do these policies reflect the 22 promise that Mr. Mosseri made in his February 2019 23 interview? 24 MR. SNEED: Object to form and 25 foundation. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 94 of 211 Page 221 1 and well-being for Instagram, and later the head of 2 youth youth safety and well-being, which evolved 3 into head of youth policy. 4 Ω. In your experience, despite what 5 the policy said, does Instagram's algorithm 6 sometimes push content that promotes suicide? 7 MR. SNEED: Object to form and 8 foundation. 9 THE WITNESS: The algorithm does 10 sometimes push content that, in aggregate, could be 11 seen as encouraging suicide but not at the 12 individual level. 13 BY MR. WARREN: 14 Ω. In your experience, despite what 15 the policy said, does Instagram's algorithm 16 sometimes push content that promotes self-injury? 17 MR. SNEED: Object to form. 18 THE WITNESS: Again, in the 19 aggregate level, the 20 21 22 Instagram might promote content that, in aggregate, could be seen as promoting self-injury, but not at the individual level. 23 BY MR. WARREN: 24 Ω. And does Instagram's algorithm push 25 that aggregated content to kids? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com 1 Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 95 of 211 Page 222 MR. SNEED: Object to form. 2 THE WITNESS: Yes. 3 BY MR. WARREN: 4 Ω. If a Meta employee let me strike 5 that question. 6 7 8 9 Is it fair to say that there's a disconnect between the suicide and self-injury policy and the aggregated content that actually gets recommended by Instagram's algorithm? 10 MR. SNEED: Object to form and 11 foundation. 12 THE WITNESS: Yes. 13 BY MR. WARREN: 14 Ω. Okay. So let's go back to right 15 around when you started at Instagram, around the 16 time of your conversation with Ms. I'm 17 going to hand you an exhibit that we will mark as 18 Exhibit 24. 19 (Document marked for identification 20 as Jayakumar Exhibit 24.) 21 MR. WARREN: It bears the Bates 22 stamp META3047MDL-040-00544758. 23 BY MR. WARREN: 24 Ω. And I'll represent to you that this 25 document was produced to the plaintiffs by Meta and 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 96 of 211 Page 259 1 A. Yes. 2 MR. WARREN: And these will be 3 Exhibits 27 and 28. 4 (Document marked for identification 5 as Jayakumar Exhibit 27.) 6 (Document marked for identification 7 as Jayakumar Exhibit 28.) 8 BY MR. WARREN: 9 Ω. And Exhibit 27 is a document 10 bearing the Bates stamp META3047MDL-037-00266408. 11 Do you have that in front of you? 12 A. Yes. 13 Ω. And what is this document? 14 A. These are the Instagram community 15 guidelines. 16 Ω. Okay. And I'll represent to you 17 this is a version of those guidelines that was 18 provided to the plaintiffs by Meta in the course of 19 this litigation. Any reason to doubt or dispute 20 that? 21 A. No. 22 Ω. And if you flip to the back of this 23 document, you'll see the metadata for it. And the 24 metadata, including the file name, indicate this 25 version of the guidelines is dated April 15, 2020. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 97 of 211 Page 260 1 Do you see that? 2 A. It says April 23, 2020. 3 Ω. Okay. And can you read the title? 4 A. Wait. Which one? 5 Ω. The title on the metadata. 6 A. Community oh. "Community 7 Guidelines 2020-04-15." Yeah. 8 Ω. So maybe there's some uncertainty 9 as to when in April, but you have no reason to 10 doubt that these are Instagram's community 11 guidelines from April 2020, correct? 12 A. Yes. 13 Ω. All right. And Exhibit 28 is a 14 15 16 17 version of the Instagram community guidelines pulled from Meta's website on January 24, 2025, as reflected in the top right corner. Any reason to doubt or dispute that? 18 A. No. 19 Ω. Okay. And if you look at the top 20 of the page, you'll see a note indicating that 21 these guidelines were moved on November 12, 2024. 22 Do you see that? 23 24 A. Ω. Yes. And the note also indicates that, 25 "None of the standards will change and they will 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 98 of 211 Page 261 1 continue to apply across Instagram." 2 Do you see that? 3 A. Yes. 4 Ω. Do you have any reason to doubt 5 that this is an accurate version of Instagram's 6 community guidelines at least as of November 12, 7 2024? 8 A. No. 9 Ω. Okay. So we'll refer to Exhibit 27 10 11 as the April 2020 version of the guidelines and Exhibit 28 as the November 2024 version of the 12 guidelines. Is that fair? 13 A. Yes. 14 Ω. Okay. Did you become familiar with 15 Instagram's community guidelines in the course of 16 your work at Meta? 17 A. Yes. 18 2. Okay. And are they a public-facing 19 document? 20 A. Yes. 21 Ω. All right. Please direct your 22 attention to the April 2020 version of the 23 guidelines. 24 Can you please read to the jury the 25 last sentence on the first page? 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 99 of 211 Page 262 1 A. 2 3 "We have zero tolerance when it comes to sharing sexual content involving minors or threatening to post intimate images of others." 4 Ω. All right. Now please direct your 5 6 attention to the November 2024 version of the guidelines, specifically the middle of the third 7 page. 8 9 10 Do you see the same sentence there? A. Yes. Ω. So at least from April 2020, to 11 November 2024, Instagram represented to the public 12 13 that it had zero tolerance when it comes to sharing sexual content involving minors? 14 MR. SNEED: Object to form. 15 THE WITNESS: Yes. 16 BY MR. WARREN: 17 Ω. And that's the right policy for the 18 company to have, zero tolerance, would you agree? 19 A. Yes, I would agree. 20 Ω. Okay. In your experience, did 21 Instagram's algorithm, in fact, sometimes recommend 22 to its users sexual content involving minors? 23 MR. SNEED: Object to form and 24 foundation. 25 THE WITNESS: Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com Case 4:22-md-03047-YGR Document 2865-1 Filed 03/20/26 Page 100 of 211 Page 263 1 BY MR. WARREN: 2 Ω. Did that include content that 3 sexualized minors? 4 A. Yes. 5 MR. SNEED: Object to form and 6 foundation. 7 BY MR. WARREN: 8 Ω. All right. I'm handing you now a 9 document that we'll mark as Exhibit 29. 10 (Document marked for identification 11 as Jayakumar Exhibit 29.) 12 BY MR. WARREN: 13 Ω. And this is a document bearing the 14 Bates stamp META3047MDL-040-00541113. 15 All right. This appears to be a 16 chat thread that includes several people, including 17 you, dated March 28, 2020. 18 Do you agree? 19 20 A. Ω. Yes. And, again, this would have been 21 very shortly after you started at Meta? 22 23 A. Ω. Yes. Did this chat take place in the 24 ordinary course of your work at Meta? 25 A. Yes. 877-370-3377 Golkow Technologies, A Veritext Division www.veritext.com