Meta's Ray-Ban Glasses Sent Videos to Humans
A privacy issue has surfaced with Meta's Ray-Ban smart glasses, which were found to be sending "sensitive" videos to human data annotators for quality assurance. The controversy highlights the growing privacy challenges for consumer-facing AI devices and the need for transparent QA processes.
The human-in-the-loop system for training Meta's AI is managed by Sama, a third-party contractor based in Nairobi, Kenya. Workers at Sama reported viewing sensitive and private footage, including people in bathrooms, changing clothes, and engaging in sexual acts, as well as clips revealing financial details like bank cards. This occurred despite Meta having systems intended to algorithmically filter out such sensitive content before human review. A key failure point is the inconsistent anonymization of the data. According to data annotators and former Meta employees, the automatic blurring of faces in the video feeds is unreliable, sometimes failing in difficult lighting conditions and leaving individuals identifiable to the human reviewers. This issue has drawn the attention of lawmakers in the European Parliament, who are now pressing for an inquiry into Meta's compliance with GDPR, especially concerning the transfer of EU citizens' data to Kenya. The contractor, Sama, has a history of controversy regarding its work for major tech firms. In 2021, the company was reported to have paid Kenyan workers as little as $1.32 to $2 per hour to label disturbing text content for OpenAI, including depictions of abuse and violence. Following exposes on worker trauma, Sama ceased its content moderation work for Meta in 2023, pivoting to focus on data annotation for computer vision. This incident is not unique in the tech industry, where human review is a common, though often opaque, part of improving AI. In 2019, Apple, Amazon, and Google all faced public backlash for similar practices involving their voice assistants. Contractors were found to be listening to audio recordings from Siri, Alexa, and Google Assistant, which sometimes included confidential information from accidental activations. In response to the 2019 privacy concerns, Apple issued a formal apology, suspended its Siri "grading" program, and later resumed it with significant changes. The company made human review an explicit opt-in for users, stopped retaining audio recordings by default, and brought the review process in-house, ending the use of third-party contractors for the task. Amazon also introduced an option for users to opt out of human review for Alexa recordings.
