Email: Re: Releasing child safety stats pre CSER v2

Raw text

Transcribed text as recorded, including headers and formatting fragments.

Message From: Tanton [/O=THEFACEBOOK/OU=EXTERNAL (FYDIBOHF25SPDLT)/CN=RECIPIENTS/CN= Sent: 9/21/2018 11:25:31 PM To: Mike Masland @fb.com]; Guy Rosen @fb.com] CC: Namrata @fb.com]; Antigone Davis @fb.com]; Vishwanath @fb.com]; Veronika Belokhvostova @fb.com] Subject: Re: Releasing child safety stats pre CSER v2 @fb.com]; Mia @fb.com]; Emily @fb.com]; Andrea Saul @fb.com]; Chris Sonderby @fb.com]; Jeff King [ @fb.com]; Scott I'm fine talking about it from a technology standpoint. One thing I heard recently is that Google also runs a lot of things on NCMEC premises and therefore have access to historical CEI data that NCMEC stores. That would greatly help our conv. featurizer work, so we should push for the same thing. Additionally, we have been worried about changing from photoDNA to another descriptor (like PDQ), but if we could rehash everything stored at NCMEC, that would alleviate that problem. Tanton From: Mike Masland Sent: Friday, September 14, 2018 2:46:33 PM To: Guy Rosen; Tanton Cc: Namrata Antigone Davis; Mia Belokhvostova; Jeff King; Scott Subject: Re: Releasing child safety stats pre CSER v2 +Tanton Emily Vishwanath Andrea Saul; Chris Sonderby; Veronika From: Guy Rosen < @fb.com> Date: Friday, September 14, 2018 at 2:33 PM To: Mike Masland @fb.com> Cc: Namrata @fb.com>, Antigone Davis Emily @fb.com>, Vishwanath Chris Sonderby < @fb.com>, Scott @fb.com>, Mia @fb.com>, @fb.com>, Andrea Saul <@fb.com>, @fb.com>, Jeff King @fb.com>, Veronika Belokhvostova < @fb.com> Subject: Re: Releasing child safety stats pre CSER v2 I think it would be super-cool if we could talk about that, if we are comfortable legally and think that tech-wise it's far enough along. This could be very good as a proof point for what modernized CEl regulation should look like in a world where Al and training data is necessary, given we are trying hard to do what is the right thing to fight child exploitation but navigating the law. Do we know this is better than what other companies are doing? (I know at least Google we've heard they just store offshore which isn't a wonderful place to be... the CEI is still stored! Have them or MS explored this approach too?) CONFIDENTIAL CONFIDENTIAL PLTF EX PLTF EX-02251 ΜΕΤΑ3047MDL-267-00000010 METANMAG-111-00000010 PLTF EX-02251_0001 On Sep 14, 2018, at 2:21 PM, Mike Masland @fb.com> wrote: Google recently announced the launch of an API to help companies identify child sexual abuse material. Maybe we could announce something similar with proactive detection + convolutional featurization? Ex: Facebook launches Al to help stop the spread of children exploitation imagery without storing it. https://www.theverge.com/2018/9/3/17814188/google-ai-child-sex-abuse-material-moderation-tool- internet-watch-foundation From: Namrata @fb.com> Date: Friday, September 14, 2018 at 8:51 AM o.com>, Mia To: Antigone Davis < @fb.com>, Vishwanath b.com>, Emily @fb.com>, Guy Rosen @fb.com>, Andrea Saul < @fb.com>, Chris Sonderby < Veronika Belokhvostova @fb.com>, Jeff King < @fb.com>, @fb.com>, Mike Masland < Cc: Scott @fb.com> @fb.com> Subject: Re: Releasing child safety stats pre CSER v2 + Scott for visibility From: Antigone Davis < b.com> Date: Monday, September 10, 2018 at 4:53 PM To: Mia @fb.com>, Namrata @fb.com>, Vishwanath @fb.com>, Emily @fb.com>, Guy Rosen @fb.com>, Andrea Saul < @fb.com>, Chris Sonderby < Veronika Belokhvostova Masland < @fb.com>, Jeff King < o.com> Subject: Re: Releasing child safety stats pre CSER v2 @fb.com>, @fb.com>, Mike The biggest challenge here is that for the most part no one thinks FB makes people suicidal but people (especially policymakers and law enforcement) do think we create the opportunity for children to be groomed. Until we have really strong mechanisms for preventing inappropriate adult/minor contact, I think we risk generating more negative press than positive - especially when we do very little to keep U13s off our platform. From: Mia @fb.com> To: Namrata Date: Monday, September 10, 2018 at 11:24 AM < @fb.com>, Vishwanath @fb.com>, Antigone Davis @fb.com>, Emily @fb.com>, Guy Rosen @fb.com>, Andrea Saul < Veronika Belokhvostova < @fb.com>, Chris Sonderby @fb.com>, Jeff King < @fb.com>, @fb.com>, Mike Masland < o.com> Subject: Re: Releasing child safety stats pre CSER v2 Ideal timing for the child safety moment is mid-late October, it needs to be before the report's release but also when we can confidently talk about these efforts externally. Nam, I'll connect with you in our sync this week to get started on the GTM here. We also just confirmed that we'll be coming to London later this month to test messaging among informed consumers. Perhaps we can tease out the CONFIDENTIAL CONFIDENTIAL ΜΕΤΑ3047MDL-267-00000011 METANMAG-111-00000011 PLTF EX-02251_0002 helpfulness of incorporating success stories here (we did something similar with suicide proactive detection and learned it was pivotal to the announcement). From: Namrata @fb.com> Date: Monday, September 10, 2018 at 7:02 AM @fb.com>, Emily @fb.com>, Guy Rosen < To: Antigone Davis < Sonderby @fb.com>, Mia @fb.com>, Vishwanath @fb.com>, Andrea Saul @fb.com>, Chris @fb.com>, Veronika Belokhvostova < @fb.com>, Mike Masland @fb.com> Subject: FW: Releasing child safety stats pre CSER v2 @fb.com>, Namrata @fb.com>, Jeff King It's great that we're thinking of announcing some of the work we're doing in this space. Specifically on a), I can work with Catherine and Mike on what we could talk about for proactive detection. When would the announcement be? On success stories, I know the messaging can be tricky, but it really clearly highlights why we need to do proactive detection for severe cases (e.g. real world harm IIC). Best, Nam From: Catherine @fb.com> Date: Monday, September 10, 2018 at 3:56 AM To: Antigone Davis < @fb.com>, Emily @fb.com>, Vishwanath @fb.com>, Guy Rosen < @fb.com>, Andrea Saul b.com>, Chris Sonderby< @fb.com>, Mia @fb.com> Cc: Namrata @fb.com>, Veronika Belokhvostova @fb.com>, Jeff King < @fb.com>, Mike Masland @fb.com> Subject: Re: Releasing child safety stats pre CSER v2 +Mike From: Antigone Davis < @fb.com> Date: Sunday, September 9, 2018 at 5:41 PM To: Emily @fb.com>, Vishwanath @fb.com>, Andrea Saul < @fb.com>, Chris Sonderby < @fb.com>, Guy Rosen @fb.com>, Mia @fb.com> Cc: Namrata b.com>, Catherine @fb.com>, Veronika @fb.com>, Jeff King < @fb.com> Belokhvostova < Subject: Re: Releasing child safety stats pre CSER v2 You've hit all the XFN initiatives: 1. Fund update of NCMEC's downloader tool 2. X-industry grooming tool + GIFCE (similar to GIFCT) 3. NCMEC VPN India CONFIDENTIAL CONFIDENTIAL ΜΕΤΑ3047MDL-267-00000012 METANMAG-111-00000012 PLTF EX-02251_0003 Ambitious to get these all done, but we will try. The XFN teams include reps from safety policy, community integrity, legal, CO, LE outreach, public policy, and comms. From: Emily @fb.com> Date: Sunday, September 9, 2018 at 1:40 PM To: Vishwanath @fb.com>, Chris Sonderby Antigone Davis < Cc: Namrata Belokhvostova < @fb.com>, Guy Rosen @fb.com>, Mia @fb.com>, Andrea Saul @fb.com>, @fb.com> @fb.com>, Catherine @fb.com>, Veronika @fb.com>, Jeff King @fb.com> Subject: Re: Releasing child safety stats pre CSER v2 Vish is correct. We'd like to package and announce the work Nam is doing to proactively detect and remove accounts that are potentially inappropriately interacting with children and a few partnership pieces that @Antigone (who I added here) has in the works, including building a download tool for NCMEC and some work with Microsoft on a potential grooming detection tool. There is a hackathon on Nov. 7-8 focused on developing this grooming tool. Related, on Oct. 29-30 in India, we will announce that we are helping to bring NCMEC VPN in country, which help train law enforcement and NGOs on protocols, etc. I'm in the early stages of learning about the partnership pieces, so Antigone please keep me honest or add things I've missed. The hope would be that we include our child nudity stats together with the package of announcements. We are planning to put together a plan around all of these pieces this week. I think we should explore (b) below but I do think it's going to be tricky given there are minors involved and I know we wouldn't want to be seen as exploiting these kids for our own benefit. From: Vishwanath @fb.com> Date: Sunday, September 9, 2018 at 12:57 AM To: Guy Rosen < @fb.com>, Andrea Saul < @fb.com>, Emily Chris Sonderby Cc: Namrata Belokhvostova < @fb.com>, @fb.com>, Mia @fb.com> @fb.com>, Catherine @fb.com>, Veronika @fb.com>, Jeff King < @fb.com> Subject: Re: Releasing child safety stats pre CSER v2 Emily and Mia have already been working on assembling all our work on child safety - product, partnerships, data to build out what a child safety moment prior to CSER would look like. We have definitely heard that people care a lot more about the real world impact of our work in this space and having something on (b) would be really be super valuable. From: Guy Rosen @fb.com> Date: Saturday, September 8, 2018 at 1:30 PM To: Andrea Saul < @fb.com>, Emily @fb.com> Cc: Vishwanath Catherine Jeff King @fb.com> @fb.com>, Chris Sonderby @fb.com>, Namrata @fb.com>, Veronika Belokhvostova Subject: Releasing child safety stats pre CSER v2 CONFIDENTIAL CONFIDENTIAL @fb.com>, @fb.com>, ΜΕΤΑ3047MDL-267-00000013 METANMAG-111-00000013 PLTF EX-02251_0004 Starting a thread on potentially releasing CEI numbers before CSER v2 to prevent the nuance being lost when we publish. This seems like a very reasonable thing because the numbers do seem scary w/o context. The one additional thought I wanted to float here is to package it with some "good news" - either: a. new product development we can talk about in the space b. I've always wanted to do this: success stories of cases we've helped children? We literally save lives (recent example here) and we have historically been worried about talking about it. I wonder if we need to lean into this more similar to how we've talked about suicide success stories. Also given some of the conversation e.g. in the UK seem to be moving from just CEI to Grooming, showing that we literally have cases where Al detected something and a child's life trajectory is changed for the better could be valuable (seriously I get goosebumps writing this). Andrea/Emily do you want to explore this? Catherine and Nam from can help you explore a+b. Chris PRIV CONFIDENTIAL CONFIDENTIAL Begin forwarded message: From: Vishwanath @fb.com> Subject: Re: Timing of CSER v2 Date: September 7, 2018 at 3:52:24 PM PDT To: Andrea Saul @fb.com>, Guy Rosen @fb.com>, Matt < @fb.com> Cc: Rachel Whetstone @whatsapp.com>, Caryn Marooney < @fb.com>, Emily p.com>, Andrea Besmehn b.com>, Debbie Frost< fb.com> Strongly agree with both the points on not using fake accounts in the context of elections as well as the child safety data. 1. We conflate the fake accounts and elections so tightly in the public domain, and then spend a lot more effort explaining why the fake accounts have nothing to do with the elections during congressional testimonies and to regulators. I feel we lose a lot of credibility in this dual stance. a. Elections Integrity is working to talk about some of the stuff pre-US Elections and I think that should help keep the "we are on it" momentum going on for Elections. 2. Child Safety is going to be hard to explain and talk about given the nuance and nature of the numbers. I fear the nuance will get lost in the large set of numbers ΜΕΤΑ3047MDL-267-00000014 METANMAG-111-00000014 PLTF EX-02251_0005 for CSER and feel talking about Child Safety earlier would really be valuable and reduce the risk of misunderstanding. From: Andrea Saul < @fb.com> Date: Friday, September 7, 2018 at 12:09 PM To: Guy Rosen < Vishwanath @fb.com>, Matt @fb.com> Cc: Rachel Whetstone < @fb.com>, Emily <dre@fb.com>, Debbie Frost Subject: Re: Timing of CSER v2 @fb.com>, @whatsapp.com>, Caryn Marooney @fb.com>, Andrea Besmehn @fb.com> My feeling is that we should release everything together post-election, with one possible exception of the CEI data. We want to have a moment where we talk about all we do in the child safety space, and given the size of the CEl removal number, I think doing it separately may better contextualize it so people aren't scared of what's happening on FB. Emily is pulling together what the narrative itself will look like, so we can see that before making any decisions - and we need approval from Mark to release CEI at all, which the data team is taking to him in the next couple of weeks for a decision. CONFIDENTIAL CONFIDENTIAL ΜΕΤΑ3047MDL-267-00000015 METANMAG-111-00000015 PLTF EX-02251_0006