The Blur Was Supposed to Be the Safeguard

This series has examined the recording-lights-and-bystander-notice">recording light and the annotators in Kenya who reviewed footage from these glasses, including intimate footage, while training Meta's AI. Automatic blurring is the safeguard Meta points to when asked how the path from capture to review can be acceptable.

Meta has never claimed that nobody outside the household sees the material. The narrower promise is that faces are stripped out before a reviewer gets near it. Without that promise, footage is simply being routed to contractors on another continent with the people in frame still identifiable.

The Guardian's reporting on 31 August described what workers say the automation delivers. Its coverage appeared alongside a report that Kmart's Anko glasses had sold out in Australia at a quarter of Meta's price. Across multiple worker accounts, the blur was applied but did not reliably hold. Intact faces reached reviewers. Credit card numbers and other identifying detail in the frame reached them too.

A physical cover can be seen, and a posted venue policy can be cited. Even a capture LED, for all its flaws, can in principle be checked by somebody other than the manufacturer. Automatic blurring offers no such check.

It runs inside a pipeline outsiders cannot inspect, on footage for which they retain no copy. The workers evaluating it are under contract not to describe most of what they see. The public knows about the failures because some workers spoke to journalists anyway. Otherwise, Meta could continue presenting the safeguard as functional while nobody outside the company had the means to confirm it.

Claims about on-device processing depend on what happens after the device has done its part. If the next step fails before human review, the earlier protection cannot carry the whole privacy promise. Nobody outside Meta can confirm that the full pipeline works. The part workers made visible did not hold up.