Meta’s Smart Glasses Spark Privacy Outcry as Workers in Nairobi

Meta’s AI Glasses Hit the Spotlight

Meta unveiled its AI‑powered smart glasses at a September 2025 event in Menlo Park, positioning the device as a hands‑free assistant that can translate languages, recognize faces, and capture moments without pulling out a phone. The launch video showed former hockey star Peter Forsberg asking the glasses who Sweden’s greatest player was, underscoring the product’s conversational interface.

The marketing copy promises “live translations, facial recognition, and a real‑time guide to the world around you,” effectively pitching the glasses as a smartphone replacement that stays on the wearer’s face. The hype is matched by a flood of ads that feature the black frames in glossy shots, but the buzz masks a deeper controversy: the glasses record audio and video continuously, feeding raw streams into Meta’s AI pipelines.

The Hidden Labor Behind the Lens

Behind the glossy launch, a subcontractor called Sama runs a data‑annotation operation in Nairobi, Kenya, where “thousands of people” label every pixel that the glasses capture. Workers draw boxes around flower pots, traffic signs, and people, then tag objects and verify quality before the data reaches Meta’s training servers. The operation sits more than 9,300 miles from Silicon Valley, on Mombasa Road, and its staff are described by a Swedish investigation as “the manual labourers of the AI revolution.”

One Nairobi annotator described seeing videos of people using the restroom or changing clothes, noting that subjects are unlikely to know they are being recorded. The same source warned that if users discovered the extent of the capture, trust in the product would evaporate. The workflow—raw video → human annotation → model training—means that every glance the wearer takes is filtered through human eyes before the AI learns from it.

Open‑Source Push for Real‑Time Privacy

In response to the privacy backlash, developers have released GlassKit, an open‑source toolkit that lets teams build and test smart‑glasses applications with built‑in privacy safeguards. GlassKit’s current demos target Rokid Glasses, providing camera and microphone capture, WebRTC streaming, a monochrome HUD, touchpad controls, and offline voice commands.

A key feature is GlassKit Eval, which runs vision regression tests across platforms without requiring a physical device. By automating checks for unwanted image capture or audio leakage, developers can catch privacy‑breaking regressions early. The toolkit also packages an “agent skill” that helps coding assistants understand the unique constraints of smart‑glasses apps—small displays, sensor access, and battery limits—so that privacy‑first logic can be baked in from the start.

Industry Implications and What to Watch

Meta’s rollout illustrates a tension that will shape the next wave of wearables: the lure of always‑on AI versus the cost of covert data harvesting. As more firms chase the “all‑in‑one” promise, the hidden labor pipeline in low‑income regions is likely to expand, raising regulatory scrutiny and public backlash.

Watch for three developments. First, any regulatory filing from the European Union or the United States that addresses continuous video capture in consumer wearables. Second, Meta’s response—whether it will publish a transparency report on annotation practices or integrate on‑device filtering to reduce raw data export. Third, adoption metrics for GlassKit or similar open‑source privacy layers; a surge in pull requests could signal that developers are taking privacy protection from a fringe concern to a core requirement.

The next few months will reveal whether smart glasses become a trusted extension of the user’s senses or remain a privacy nightmare built on unseen hands.