The Human Review in Kenya's Smart-Glasses Investigation

Investigations by Svenska Dagbladet and Göteborgs-Posten reported that annotators working for Sama in Nairobi reviewed footage captured by smart glasses. The material reportedly included people undressing, using bathrooms, having sex and entering card details.

The people shown had no reason to expect a worker in another country might inspect those moments. Distance does not soften the breach. It makes the route from capture to review harder for the subject to see.

Former Meta employees told reporters that faces are blurred automatically. Annotators described cases in which anonymity failed and faces remained visible. An automated filter cannot settle the privacy question before a human sees the material if nobody can show when it worked or failed.

A safeguard needs an audit trail showing when it ran and failed, who could access the original material, and what happened after an exception. Without one, a bystander has no reliable way to understand where a moment captured by glasses can travel.

Kenya's Office of the Data Protection Commissioner opened an investigation after a petition from The Oversight Lab, according to The Verge and Techweez. The story brings a neglected part of wearable AI into view. People who label or review data are part of the product's privacy model.

For smart-glasses makers, the question extends past what a wearer can record. It includes where footage goes, who can view it and whether the person caught on camera can ever learn that review happened.

The reported workflow separates collection from review by geography and role. That can make the route difficult to understand, especially when the person in the footage has no account with the service and no relationship with the reviewer. A privacy description that names only the wearer-facing feature leaves out part of the processing chain.

Human review is a control point. Blurring, access permissions, retention limits and escalation procedures may reduce exposure, but each has to operate before and during review. The investigation does not establish the policy for every product. It does show why review cannot be treated as an invisible backend detail.