Ronan Bradley: "The most significant risk, if this leaks, is that we will face significant policy pressure to use techniques like this... to enforce age restrictions"

"This is a summary of the age affinity data ask from messenger which we discussed yesterday. The ask is to assist with resolving the escalation." "The specific business need is to provide additional age related information to allow us to build and validate high quality age models in key countries outside of the US (where the birthday post related information proved sufficient). We have a list of 11 hi-pri countries (BR, MX, FR, DE, Poland, ID, IN, TH, EG, NG, KR) which we want to focus on in order to give us global coverage." "We conducted initial explorations on the messenger sampled data set and found that there was a significant amount of age related information. We would like to get permission to extract and review age related historic messenger data. The extraction will be performed algorithmically. However, it is also necessary to review messages as a quality check to ensure that the algorithm is extracting accurate age information (ie Happy 16th birthday) of sufficient quality to allow the building and validation of the age affinity model." "The policy team has summarized the following risks:" "The most significant risk, if this leaks, is that we will face significant policy pressure to use techniques like this, even if imperfect, to enforce age restrictions on Facebook & Messenger. If the pressure got high enough, which is possible, we might have to pro-actively try to identify and then remove everyone under 13 from Facebook & Messenger. (Starting in May next year this will be under 16 in many parts of the EU)." "People and policy makers are particularly concerned about Messenger content because of the perception that these are 1:1 communications that they don't expect to be 'read' by Facebook. From the policy point of view, in-product experiences (Messenger Suggestions) and safety/anti-terrorism efforts are the easiest uses to explain and defend. Research like this project, especially research into teens creates policy risk as regulators might decide that this was one step too far and then try to significantly limit the research we can do. This project creates increased risk above the usual Messenger research we do for 3 reasons:" "(1) The 2 main privacy protections we rely on to defend doing research on Messenger content is that (1) it's only done on a small random sample (we currently use a 3% sample) and (2) we take steps to de-identify the content before any human review. The proposed analysis pushes on both of those protections because it requires both the full Messenger corpus and pairing the content with profile photos for human reviewers." "(2) It's about teens who are especially sensitive in relation to Facebook right now." "(3) And (3) because a marketing insights deck with research into teens emotions was leaked in Australia and the launch of Messenger Suggestions in Spain, we are in a period where we are more likely than usual to have to answer broad questions about what research we do on minors and on Messenger." "While we have had extensive discussion with a number of people in messenger policy/legal/eng/analytics teams, I do not believe that it has yet been escalated to David Marcus and that would be an obvious next step for you to help with."