Email: Re: Questions on eating disorder detection / enforcement

Raw text

Transcribed text as recorded, including headers and formatting fragments.

Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 1 of 19 AMENDED Exhibit 179 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 Case 4:22-md-03047-YGR From: Sent: To: Document 2757-7 Filed 02/20/26 Page 2 of 19 [/O=THEFACEBOOK/OU=EXCHANGE ADMINISTRATIVE GROUP (FYDIBOHF23SPDLT)/CN=RECIPIENTS/CN Thur 9/17/2020 4:56:06 PM (UTC) Miki Rothschild Cc: Newton Liza Crenshaw ]; Yoav Shapira ; Vaishnavi Subject: Re: Questions on eating disorder detection / enforcement PRIV From: Miki Rothschild Date: Thursday, September 17, 2020 at 12:42 PM To: Cc: Karina Newton Vaishnavi J >, Yoav Shapira Liza Crenshaw Subject: Re: Questions on eating disorder detection / enforcement From: Date: Thursday, September 17, 2020 at 9:37 AM To: Miki Rothschild Karina Newton Vaishnavi J PRIV Yoav Shapira Karina Cc: Stephanie Otway <stephanieotway@fb.com>, Tara Hopkins <tarahopkins@fb.com>, Rachael Gallagher HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246746 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Liza Crenshaw Subject: Re: Questions on eating disorder detection / enforcement PRIV From: Miki Rothschild Date: Thursday, September 17, 2020 at 9:32 AM To: Shapira Cc: Liza Crenshaw Vaishnavi J Subject: Re: Questions on eating disorder detection / enforcement PRIV From: Date: Thursday, September 17, 2020 at 9:17 AM To: Vaishnavi J Cc: Liza Crenshaw Karina Newton Yoav Shapira Subject: Re: Questions on eating disorder detection / enforcement HIGHLY CONFIDENTIAL (COMPETITOR) Miki Rothschild Page 3 of 19 Karina Newton Yoav META3047MDL-031-00246747 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 4 of 19 From: Date: Thursday, September 17, 2020 at 11:54 AM To: Karina Newton Cc: Crenshaw Yoav Shapira Miki Rothschild Subject: Re: Questions on eating disorder detection / enforcement Vaishnavi J PRIV Liza HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246748 Thanks, Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 5 of 19 PRIV From: Karina Newton Sent: Thursday, September 17, 2020 10:49 AM To: Cc: Miki Rothschild Subject: Re: Questions on eating disorder detection / enforcement A/C Priv Karina Yoav Shapira >; Vaishnavi J I seek your advice and counsel. PRIV HIGHLY CONFIDENTIAL (COMPETITOR) Liza Crenshaw META3047MDL-031-00246749 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 6 of 19 From: Date: Thursday, September 17, 2020 at 7:07 AM To: Cc: Crenshaw Yoav Shapira Vaishnavi J Karina Newton Miki Rothschild Liza Subject: Re: Questions on eating disorder detection / enforcement Thanks for explaining, Appreciate that it is a lot of work [and that I was over-simplifying!] but I do think this is incredibly important reputationally - for the company and Adam. The Telegraph article is one piece but most crucial is the inquest into Molly Russell's death, which we expect to happen soon [pre-inquest hearing next week]. If we are in scope which is very likely, the focus will be on what we were recommending Molly - and most crucially whether we have done enough to prevent 'future deaths'. On explore and recommendations in feed we're exposed but have made the changes promised and are now behind due to classifiers + Covid. On search we're exposed with nowhere to hide. We're not necessarily saying blocking users in this scenario - but stopping anything from appearing in search when you type things like proana, thinspo etc. [i.e. anything that could be inherently violating]. Happy to join a call if it's easier to discuss it live at any point. Thank you! From: Date: Thursday, 17 September 2020 at 12:50 To: Cc: Crenshaw Yoav Shapira Vaishnavi J Karina Newton Miki Rothschild Liza Subject: Re: Questions on eating disorder detection / enforcement > Apologies if this is over-simplifying but could we - as a first step at least - apply the same list of blocked terms that we do to hashtag search to ALL search. The problem is that it's not a simple blocklist in hashtag search. It's a complicated system that tries to match prefixes of hashtags against blocked hashtags at runtime. This would take significant effort to port over to user search, and would add a lot of runtime overhead and complexity for something that would be very brittle (because of the performance impact, it's limited to prefixes of length 6-8, "proana" just happens to fit in that window. We wouldn't even know if it would be possible to run a useful set of prefix sizes in user search with reasonable performance until we implement and test it). Also, it's not clear that this would work in user search - do we have a user named HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246750 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 7 of 19 "proana" in Tier 1? I don't think we want to get into the business of blocking users based on matching their username with hashtags. From: Date: Thursday, September 17, 2020 at 6:29 AM To: Yoav Shapira Cc: Crenshaw Vaishnavi J Miki Rot Miki Rothschild Karina Newton Liza Subject: Re: Questions on eating disorder detection / enforcement We did broadly say on background that we're exploring what more we can do on account search. @Yoav Overall he thinks too much is being recommended across the board but acknowledged it's good certain terms can't be found in hashtag search [we also blocked some of the ones he flagged, which helped]. But 'tags' is the 3rd tab along so when you search proana [for example] loads of stuff comes up - the vast majority violating - already in the first two tabs [Top and Accounts], even if there's nothing in the 3rd tab. He thinks that is us not delivering on the promise made by Adam in February 2019 that SSI content won't show up in search. Apologies if this is over-simplifying but could we - as a first step at least - apply the same list of blocked terms that we do to hashtag search to ALL search. So if I type in proana, nothing appears. If people type in exact account names they could find them but we would not be serving them up. He's holding us to the fact that Adam said we wouldn't show this stuff in search and yet, because of the disparity between hashtag search and account search that's hard to explain, it's incredibly easy to find. Worth noting he will also point to this stuff still being recommended in Explore and suggested accounts in feed too though it's easier for us to lean on the classifiers and covid there [i.e. we are doing what Adam said and when we know about more we can stop more being recommended]. Search is harder because of the disparity between hashtags and accounts. Hope that helps! From: Yoav Shapira <yshapira@fb.com> Date: Wednesday, 16 September 2020 at 21:20 To: Cc: Vaishnavi J HIGHLY CONFIDENTIAL (COMPETITOR) Karina Newton META3047MDL-031-00246751 Case 4:22-md-03047-YGR Crenshaw Document 2757-7 Filed 02/20/26 Page 8 of 19 Liza Miki Rothschild Subject: Re: Questions on eating disorder detection / enforcement IMO, I think we can say that, sure. But no promises on actual changes or a timeline. We really are exploring organizational options to get Search Integrity in a better place. @ did the reporter buy that Hashtags are better than Search? So we have reason to believe if we implement the same measures on Search, will be OK? Or did they still point at content on Hashtags, Explore, Feed, etc? Yoav From: Date: Wednesday, September 16, 2020 at 12:45 To: Cc: Crenshaw Miki Rothschild Vaishnavi J Karina >, Yoav Shapira Liza Subject: Re: Questions on eating disorder detection / enforcement Coming in blind, but any chance we can say we're exploring addressing the search vs hashtag gap, or are we not? From: Date: Wednesday, September 16, 2020 at 11:44 AM To: ப Cc: Crenshaw Yoav Shapira Miki Rothschild Subject: Re: Questions on eating disorder detection / enforcement + @ Liza Wanted to send a quick update as we did a phone briefing with the Telegraph reporter writing the piece that sparked this thread this afternoon. HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246752 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 9 of 19 As expected, he was very focused on the content we recommend and concerned that violating self-harm and eating disorder content is still too easy to find as a result. He focused specifically on the fact we recommend accounts in search and was confused as to why we do that given Adam said last year that SSI content would 'no longer show up in search, hashtags or account recommendations'. I know internally we have different ideologies for search vs hashtag or recommendation surfaces but externally all are seen as places where we serve content, so the argument doesn't fly. did an excellent job of managing the question and pivoted to talking about classifiers and impact of covid [i.e. if we can find and remove more violating content it won't be there at all], but we are definitely exposed on this and it's a risk. Let me know if you have any questions or I can provide more info to help with next steps. From: Date: Wednesday, 16 September 2020 at 11:14 To: Shapira Cc: , Karina Newton Liza Crenshaw Subject: Re: Questions on eating disorder detection / enforcement + Miki Rothschild Vaishnavi J >, Yoav That's great, thank you! The main phrases that spring to mind are inherently violating terms like 'proana' and 'thinspo' - and ideally all the other words/phrases on our blocked hashtags list, which we can share with you. Happy to continue this offline. From: Date: Tuesday, 15 September 2020 at 23:13 To: Karina Newton Vaishnavi J Yoav Shapira Cc: Liza Crenshaw Subject: Re: Questions on eating disorder detection / enforcement HIGHLY CONFIDENTIAL (COMPETITOR) Miki Rothschild META3047MDL-031-00246753 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 10 of 19 Hi While we discuss potential solutions with an inform module, I want to offer help with an interim solve to drive parity between hashtags and accounts. It would still be good to discuss blocking selected ED-related search terms for accounts. As others mentioned, it's hard to defend why we'd block certain terms in hashtag search and not in account search, and it's easy to find violating ED accounts in a couple of taps. When I search for 'thinspo' or 'proana' the accounts shown are overwhelmingly violating. Do you have a list of 'violating' phrases that are being actioned for hashtags and NOT being actioned for accounts? i.e what phrase should be included in an account's name to be considered inherently violating? We can tier those accounts in the meantime to limit their discovery on Search. We can probably take this discussion offline in a chat thread with Karina. Product Manager - Instagram Wellbeing C: From: Date: Tuesday, September 15, 2020 at 4:00 PM To Cc: Vaishnavi J Karina Newton Yoav Shapira Liza Crenshaw Subject: Re: Questions on eating disorder detection / enforcement + Miki Rothschild Search's perspective - To build the inform module for SSI queries, (like we did for covid and vaccines), it would take some time, plus tradeoff against search quality/shopping search goals for H2, as our eng are currently committed to those workstreams. In the past, to build the inform module, the search team has utilized the regex(search terms to trigger this module), content, url (translated, policy approved and localized) in the inform module, that FB's health integrity team provided, so we would need this from the FB SSI team, before we can build the module on IG Search. Additionally - I do want to call out that post launching the inform module for both covid and vaccines, this was not seen as a sufficient treatment, to solve for the risk of the accounts being discovered on search - as users can still discover bad vaccine accounts for vaccine related queries. In the case of covid, we relied on account tiering to add bad covid accounts to Tier 1 or 2 and in the case of vaccines, the tiering team is still working on this with FB Health Integrity, so we should parallelly plan for an account tiering solve for this issue. HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246754 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 @Karina - Please add Page 11 of 19 and me to the discussion/meeting and we can work on figuring out a solution and timing here. Thanks, From: Sent: Tuesday, September 15, 2020 3:20 PM To: Vaishnavi J Yoav Shapira Cc: Karina Newton <alimcinerney@fb.com>; Claire Lerner <clerner@fb.com>; Liza Crenshaw <lizacrenshaw@fb.com>; Miki Rothschild Subject: Re: Questions on eating disorder detection / enforcement Agreed that an inform module for ED search terms would be really helpful – thank you! Any sense of how long that would take? Would you be comfortable with us confirming on background tomorrow that 'we're exploring ways to offer more support to people who search for these types of accounts' or way too premature? It would still be good to discuss blocking selected ED-related search terms for accounts. As others mentioned, it's hard to defend why we'd block certain terms in hashtag search and not in account search, and it's easy to find violating ED accounts in a couple of taps. When I search for 'thinspo' or 'proana' the accounts shown are overwhelmingly violating. Finally - on my original #3 - it sounds like even when we do launch ED classifiers, they won't be able to run on account elements, so wouldn't pick up accounts based on 'proana' in the name/bio, for example. Does anyone have more details on that? I think the rationale is GDPR-related, but is coming from privacy policy at a global level. Thanks again for all the help! From: Vaishnavi J Date: Tuesday, 15 September 2020 at 18:31 To: Karina Newton >, Yoav Shapira Cc: Liza Crenshaw Miki Rothschild Subject: Re: Questions on eating disorder detection / enforcement An inform module seems like the best and most thoughtful mitigation - we've in parallel been working to add more ED organisations to the list of organisations that are turned up in the SSI interstitial, and Kira and the policy team have been adding to a list of groups. Happy on my end to prioritise gathering more of these groups for inclusion. From: Karina Newton HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246755 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 12 of 19 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246756 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 13 of 19 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246757 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 14 of 19 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246758 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 15 of 19 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246759 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 16 of 19 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246760 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Page 17 of 19 HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246761 Case 4:22-md-03047-YGR Document 2757-7 Filed 02/20/26 Ο Page 18 of 19 We're concerned as this means these accounts - which are clearly violating - will remain easy to find and may continue to be recommended - both to vulnerable people, and to journalists who are looking to write stories. • Even though we couldn't do it in the EU any time soon, being able to use classifiers to find violating accounts elsewhere in the world would still be helpful, as a) it will help with our eventual discussions with the IDPC, to show that ROW has technology that we don't have in the EU; and b) people in the EU still see accounts from ROW, and there would be fewer accounts promoting eating disorders coming from ROW. We're concerned that our ability to find, remove and stop recommending these pro-ED accounts is going to be limited by #1 and #3, and we're going to keep getting negative press cycles about how we're failing to fulfil the promises we've made. It's also worth noting that the Molly Russell inquest will likely focus on the content we were recommending her. While that will be about suicide/self-harm/depressing content more than eating disorder content - showing we've made improvements to avoid recommending harmful content will be key. @ @Vaishnavi J - would love your thoughts on #1 - whether you agree and what might be possible? @ @ @ what are your thoughts on #3? What discussions do you think we should have internally, given this appears to be driven by GDPR concerns? and I chatted earlier, and noted that we shouldn't bring ED classifiers into the IDPC discussions yet, as we're on thin ice as it is! I realise the above is A LOT, so to jump on a call to discuss any of this next week. Have a good weekend! Thanks HIGHLY CONFIDENTIAL (COMPETITOR) META3047MDL-031-00246762 Case 4:22-md-03047-BOCONMENT SLIP SHEET Pocument 02/20/26 Page 19 of 19 Bates Number: META3047MDL-031-00246746-META3047MDL-031-00246762 BEGATTACH: META3047MDL-031-00246746 ENDATTACH: META3047MDL-031-00246762 TNAG BEGBATES: TNAG ENDBATES: Create Date: 9/17/2020 9:56 AM Last Modified Date: 9/17/2020 9:56 AM Custodian: CRENSHAW LIZA; JAYAKUMAR_VAISHNAVI; NEWTON_KARINA; ; ROTHSCHILD_MIKI; Title: Author: Confidentiality Designation: META Pointer: META Pointee: