A new investigation has uncovered claims that human workers in Kenya are regularly viewing intimate and sensitive footage captured by users of Meta’s AI-powered Ray-Ban smart glasses.
The report, led by Swedish outlets Svenska Dagbladet and Goteborgs-Posten, reveals the hidden role of data annotators far from the device user in the technology’s training process.
Sold as a wearable AI assistant capable of translating languages, describing surroundings, snapping hands-free photos and videos, and answering questions about what wearers see, Meta’s Ray-Ban smart glasses rely on cloud-connected processing and extensive training data.
Yet footage recorded by users sometimes without their full awareness is reportedly sent to Nairobi, Kenya, where contractors employed by outsourcing firm Sama open and label it to help improve the AI’s performance.
Workers told investigators they often encounter material far more personal than everyday scenes, along with accidentally captured financial information like bank cards.
Read also: Mark Zuckerberg takes stand in U.S. trial accusing Meta of youth social media addiction
One worker told reporters that users likely don’t know just how much the glasses can record once they are activated.
According to the investigation, retail staff in some European markets have given inconsistent or misleading answers about whether video data stays on the device or is uploaded to Meta’s servers.
Independent tests cited in the report suggest many of the AI functions require remote processing, meaning video and audio are transmitted away from the device.
Sama formerly Samasource provides data annotation services for major tech clients including Meta and OpenAI.
Its workers are bound by strict confidentiality agreements and often earn low wages while facing heavy workloads.
Meta’s privacy policies state that user content can be reviewed by humans to improve its products and ensure safety, and for European users the company says its Irish subsidiary oversees compliance with the EU’s General Data Protection Regulation (GDPR).
Still, the investigation has raised questions about cross-border data transfers, especially to countries like Kenya that lack an EU adequacy decision on data protection.
While Meta has defended its privacy practices in past statements, the workers’ accounts suggest that the boundary between automated AI processing and human-involved review may be much less clear and more intrusive than many users expected.
