The glasses, created in partnership with eyewear giant EssilorLuxottica and sold as Meta Ray-Ban smart glasses, are promoted as an AI assistant capable of translating languages, answering questions and helping users interpret what they see. During the product launch in Menlo Park, Meta founder Mark Zuckerberg described the technology as a glimpse of the company’s future.
But the investigation suggests the reality behind the product is far more complex.
Workers reviewing private footage
Thousands of data workers in Nairobi, Kenya, help train the AI used by the glasses. Many are employed by outsourcing firm Sama, which provides data annotation services for tech companies.
These workers review images, videos and conversations generated by the devices so the AI can better recognize objects and understand user requests.
According to several employees interviewed by the newspapers, the material they see often contains deeply private scenes. Workers described reviewing clips that appear to come from homes in Western countries, including footage of people using bathrooms, undressing or engaging in sexual activity.
Others said the recordings sometimes show bank cards, private conversations or people watching explicit content while wearing the glasses.
Workers are bound by strict confidentiality agreements and cannot bring recording devices into the offices where they review the data.
Sales staff told reporters that users can decide whether their data is shared with Meta. However, technical analysis carried out by journalists suggested that the glasses regularly communicate with Meta servers in Sweden and Denmark when the AI features are used.
According to Meta’s own terms of service, interactions with its AI can be reviewed automatically or manually by humans. Voice, text, images and sometimes video must be processed through Meta infrastructure for the assistant to function. That means the data cannot be handled only on a user’s phone.
Privacy experts say many users may not realize how much data is collected. Kleanthi Sardeli, a data protection lawyer at the Vienna-based group None Of Your Business, said the technology raises transparency concerns, especially if users do not realize when recording begins.
“Once the material has been fed into the models, the user in practice loses control over how it is used,” she said.
Meta’s response
Meta says data from the glasses can be transferred and processed globally, including by partners and service providers. Under European law, the company remains responsible for protecting user data even if it is handled outside the European Union.
The company declined to answer detailed questions from the newspapers about how private material may reach subcontractors such as Sama.
A spokesperson said only that media used by the AI assistant are processed according to Meta’s terms of service and privacy policy.
Earlier, Qazinform News Agency reported that Meta and Advanced Micro Devices signed a multi-year, multi-generation agreement to deploy up to 6 gigawatts of AMD Instinct GPUs to power Meta’s next generation of AI infrastructure.