A Swedish investigation report has revealed that Meta AI smart glasses are sending users’ private recordings to human reviewers at a third-party company in Kenya.
According to a report by Swedish publication Svenska Dagbladet, the glasses, which feature cameras, microphones, and an AI assistant activated by saying “Hey Meta,” are advertised as safe and privacy-focused.
However, the investigation reveals that when users ask the glasses’ AI questions or sometimes even when they simply record footage, it is automatically sent to Meta’s servers for processing.
From there, the information ends up in the hands of human data annotators, and the function cannot be turned off.
Workers Say They See Extremely Private Footage
The investigation found that Kenyan contract workers are exposed to footage that many users would never expect others to see. Workers interviewed said they routinely viewed clips of:
- People using the bathroom
- Individuals getting undressed
- Intimate moments.
- Bank cards, passwords, and financial details
- Users watching adult content
One worker said, “In some videos you can see someone going to the toilet or getting undressed.” Another said, “I don’t think they know, because if they knew, they wouldn’t be recording.”
Also Read: Meta Introduces New Feature to Identify AI-Generated Images
Other workers described situations where glasses were left on a bedside table and continued recording, capturing someone’s partner walking in and changing clothes without realizing she was being filmed.
Several employees also said they felt pressured to review this sensitive content or risk losing their jobs.
Why Meta AI Sends The Footage to Kenya
Meta trains its AI systems by having real people, called annotators, watch videos and images and label what they see. These workers point out things like objects, actions, places, or anything else the AI needs to learn to recognize.
This kind of human review is common across the tech industry, as AI models don’t automatically understand what they’re looking at unless theyare trained.
However, what’s unusual in Meta’s case is that some of the material being reviewed reportedly comes from inside users’ homes, which raises serious privacy concerns.
Also Read: Meta Picks Safaricom for Undersea Internet Cable Connecting Oman and Kenya
To protect users, Meta uses automated tools that are supposed to blur faces or hide anything that might identify a person. But reports say these tools sometimes fail.
When that happens, the workers still end up seeing clear, personal, or very private clips, often from users in Europe and other regions, according to the report.
This means the human reviewers may unintentionally watch moments that people never expected anyone else to see, stressing ongoing worries about how private data is handled when training powerful AI systems.
Meta Responds to Privacy Concerns
Meta’s privacy policy does warn that data, including conversations, audio, text, and images, may be reviewed by automated systems or humans. It also cautions users: “Do not share information that you don’t want the AIs to use and retain.”
In response to media questions, a Meta spokesperson said the company sometimes uses contractors to review data “to improve people’s experience with the glasses,” and insisted that this is disclosed to users. They added that the data is “filtered to protect people’s privacy.”
However, the Swedish investigation suggests that in practice, the filtering does not prevent exposure of intimate or highly identifiable moments.
Follow our WhatsApp Channel and X Account for real-time news updates.
