The 'data annotators' behind Meta's smart glasses can see all of your private videos and photos
A joint investigation by Swedish media outlet Svenska Dagbladet uncovered how Meta's AI-powered smart glasses are being manually processed by workers in Kenya, sending highly sensitive user data to the tech giant's systems. Meta's smart glasses are marketed as a convenient assistant for everyday life, but behind the scenes, despite promises of privacy, vast amounts of sensitive information are being sent to the tech giant's systems.
Meta’s AI Smart Glasses and Data Privacy Concerns: Workers Say “We See Everything”
Mark Zuckerberg of Meta unveiled the Meta Ray-Ban Glasses as a future-defining product, with features like real-time translation and object recognition, designed to be a powerful assistant that rivals smartphones.
Meta announces the second generation model of AI glasses 'Ray-Ban Meta' and 'Meta Ray-Ban Display' - GIGAZINE
But behind the scenes of this innovation lie thousands of workers far from Silicon Valley, in Nairobi, Kenya.
Meta works for a company called Sama, where workers called data annotators manually identify and label images and videos to train AI systems. These workers, who are essentially manual laborers in low-income countries, are supporting the AI revolution and are under non-disclosure agreements.
According to a study by Svenska Dagbladet, data annotators in Kenya are exposed to highly intimate footage of users' everyday lives, including scenes of them using the toilet, undressing, naked partners, and even sexual intercourse. One annotator said users are exposing their naked bodies and private lives without realizing they are being recorded.
The data being processed includes not only video but also audio and text data. Annotators may also review chats containing bank card details, criminal activity, protests, or sexual comments. One data annotator said that if these contents were leaked, it would cause a major scandal.
Sales of smart glasses are booming, tripling from 2 million units sold in 2023 and 2024 to 7 million units in 2025. Svenska Dagbladet reports that a major Swedish retailer has found cases where store clerks have given customers false or unclear answers about data sharing, such as 'everything is stored locally within the app' and 'nothing is shared with Meta.'
A Svenska Dagbladet reporter actually purchased the device and analyzed its network traffic, finding that its AI features required an internet connection and that it frequently communicated with Meta's servers in Sweden, contradicting the store clerk's claim that the device was only processed locally.
by cavebear42
Meta's terms of service state that data, including audio, images, and video, may be processed to improve its AI and may be subject to manual review by humans. Users must agree to these terms in order to use the AI features, and details about what data is stored, for how long, and who has access to it remain unclear.
Experts point out that users may not be fully aware that their data is being used to train AI. In Europe, in particular, the EU General Data Protection Regulation (GDPR) requires strict transparency regarding the location of data processing, but there are doubts as to whether data transfers to third countries such as Kenya meet an adequate level of protection.
Furthermore, the anonymization technology used to protect privacy appears to be flawed. A former Meta employee said that the algorithm that automatically blurs faces can fail in poor lighting conditions, but that Kenyan data annotators sometimes review footage with faces and bodies clearly visible.
When Svenska Dagbladet contacted Meta about the issue, they did not directly respond to specific questions, other than to say that data processing is carried out in accordance with their terms of use and privacy policy. Sama, the company that handles the data, did not respond.
Related Posts:
