Meta Ray-Bans send ‘sensitive’ videos to human data annotators

A new report says that video feeds from Meta Ray-Ban smart glasses are sent for review by human data annotators in Kenya, and that the footage includes sensitive content that is supposed to be excluded.

Whistleblowers says that the video seen by third-party contractors used by Meta includes everything from people having sex to bank cards …

Meta Ray-Ban smart glasses can capture video in two ways. First, you can activate video recording manually in order to capture point of view footage. This can be a great hands-free way to record experiences like a roller coaster ride, as well as incidents that might occur while driving or cycling.

Second, you can use the AI feature to ask questions about whatever you are looking at through the glasses. It’s well understood that this AI processing is handled on Meta’s servers and therefore that video footage needs to be sent to these for analysis.

However, a report by Swedish site SVD says that footage is sent to human data annotators whose job it is to manually identify objects seen in these clips. A worker from a third-party contractor based in Kenya says that this footage sometimes includes very sensitive content.

The workers in Kenya say that it feels uncomfortable to go to work. They tell us about deeply private video clips, which appear to come straight out of Western homes, from people who use the glasses in their everyday lives. Several describe video material showing bathroom visits, sex and other intimate moments […]

“Someone may have been walking around with the glasses, or happened to be wearing them, and then the person’s partner was in the bathroom, or they had just come out naked”, an employee says.

The circumstances in which these sensitive videos are captured is very unclear from the report. For example, there is reference to people wearing the glasses while having sex, which would appear to be a very deliberate use. However, this would also seem to indicate that video footage is sent for review even when someone is manually recording rather than using Meta AI.

There is definitely a lack of transparency about what footage is sent to Meta when using the AI function. For example, if you look at a car and ask Meta to identify the make and model, at what point does it cease sending footage? Is it five seconds later, 10 seconds, 30 seconds? Is it as soon as the question has been answered, or does it continue recording in case you ask further questions?

The company’s own terms of use are exceedingly vague.

The terms state that “in some cases, Meta will review your interactions with AIs, including the content of your conversations with or messages to AIs, and this review can be automated or manual (human).”

SVP says that when it asked Meta for details, the company simply referred them back to the Terms of Service and Privacy Policy.

Frustratingly, the report says that the site analyzed the network traffic to see what was being sent but then provides absolutely no insight.

When we then analyse the network traffic from the app, we see that the phone has frequent contact with Meta servers in Luleå, Swden, and Denmark.

Former Meta employees say that sensitive data isn’t supposed to be sent for human review, but this relies on algorithmic identification of that sensitive data, which isn’t always successful.

9to5Mac’s Take

I use the glasses myself. The novelty of the AI feature wore off rather quickly, but they’re a very convenient way to shoot hands-free POV footage. Although I’d never use them to shoot anything sensitive, I would be pretty outraged to discover that Meta is capturing manual video recordings.

The report is frustratingly lacking in hard information, but I guess it serves as a reminder to use any AI service with caution when it comes to sensitive data of any kind – or any Meta product.

Photo: 9to5Mac/Ray-Ban