When you wear Ray-Ban smart glasses and call on Meta AI to help you identify objects in front of you, you probably don't realize that these first-person perspective images are being closely watched by "humans" far away in Africa. According to a joint investigation by Swedish media and revelations from outsourced employees in Kenya, Meta is sending a large amount of daily video footage recorded by Ray-Ban smart glasses to data labeling personnel in Kenya for review and training. Shockingly, due to frequent malfunctions of the automatic decryption system, these outsourced employees are forced to view extremely private Western family footage, including scenes of using the toilet, nudity, sexual activity, and even bank cards.
The working class in the AI revolution: Kenyan "data labelers"
According toA joint investigation by Swedish media outlets Svenska Dagbladet and Göteborgs-PostenIt is pointed out that when Ray-Ban smart glasses users activate the AI function (such as asking questions like "Hey Meta, look at what's in front of me and tell me..."), the glasses will capture images and videos and send them back to Meta's servers for processing.
However, the journey of this data did not end there. To continuously train and optimize Meta's AI visual recognition system, these images, containing real-life scenes of users, were continuously fed to Sama, an outsourcing company located in Nairobi, the capital of Kenya. Here, thousands of workers, known as "data annotators," spend their days sitting in front of screens, drawing frames and labeling objects in these images.
Algorithm-based decryption fails, leaving users' lives completely "naked".
The most serious problem is that Meta is overconfident in its automatic privacy masking technology.
Several Sama employees who signed strict confidentiality agreements revealed to the media that they were frequently forced to review extremely private footage. This is because the system's built-in "automatic decryption algorithm" often fails in complex lighting conditions, fast-moving scenes, or atypical scenarios. This forces annotators to directly witness users using the toilet, exposing their bodies, and even engaging in sexual activity; furthermore, highly sensitive personal financial information, including bank card numbers and passwords, is also presented to these outsourced employees without any protection.
A Sama employee stated, "We can see everything, from the living room to nudity. If users knew exactly what they were recording, they would never dare to use these glasses again."
Meta's cold treatment and the "get-out-of-jail-free card" hidden in the terms and conditions
Faced with such serious privacy allegations, Meta's response appeared evasive. The company did not directly address specific preventative measures against the leak of private videos, only emphasizing that it "strictly adheres to the terms of service in handling data."
In fact, if you carefully read Meta's lengthy terms of service, which almost no one reads to the end, you'll find that Meta has already subtly included a statement: "Some AI interactions may be subject to human review." Vienna-based data protection lawyer Kleanthi Sardeli pointed out that this practice severely lacks transparency under EU regulations (such as the GDPR). This is because when users wake up the AI assistant with their voice, most people are completely unaware that the camera on their glasses is recording, let alone knowing that this footage will be sent to the other side of the world for strangers to view.
Analysis of viewpoints
While smartphones do collect data, they are mostly kept in a pocket or on a table; smart glasses, on the other hand, are worn on the face, and the world they see is the world you see with your own eyes. When you walk into the bathroom, bedroom, or even open your wallet, as long as the AI is on standby or accidentally activated, your life becomes free training material for tech giants.
This incident also exposes an open secret in the tech world: all currently claimed advanced AI vision models are extremely reliant on human "labeling." Algorithms are not yet smart enough to understand the world on their own; they need Kenyan workers to tell them "this is a cup" or "that's a cat."
However, in this process, technology companies often sacrifice user privacy in pursuit of rapid model iteration. In the future, ensuring the purity and anonymity of cloud-based AI training data before "on-device AI" technology matures will be a serious issue that giants like Meta, Google, and Apple, dedicated to developing smart glasses, must clearly address.
