Over 1,100 Kenyan data annotation workers at Sama's Nairobi office lost their jobs in April after Meta terminated its outsourcing contract, a termination that came directly after those workers disclosed to journalists that their queues included intimate footage captured without consent by Meta's Ray-Ban smart glasses.
The sequence of events is documented and sequential enough to require no inference. In February 2026, a joint investigation by two Swedish publications published accounts from Sama workers in Nairobi describing what they were being paid to review. The footage came from Meta's Ray-Ban smart glasses, which sold more than seven million units in 2025, tripling prior-year volume. Users wear them as ordinary glasses. The cameras capture first-person video that can be submitted to Meta's AI training pipeline. Workers described queues containing footage of people in bathrooms, footage of sexual activity that the other parties clearly had not consented to, footage of children in private moments, and banking information visible in what should have been routine daily footage. This was not described as an occasional edge case. Workers said this kind of content appeared regularly in their assigned annotation tasks.
Meta paused its work with Sama shortly after the investigation published. On April 16, Sama issued formal redundancy notices to 1,108 employees, giving them six days to clear their desks. Severance was the Kenyan statutory minimum under Section 40 of the Employment Act: 15 days of pay per completed year of service. A Meta spokesperson's explanation for ending the contract was notably constructed. The company said it had decided to end work with Sama "because they don't meet our standards." Sama disputes this framing entirely. Naftali Wambalo of the Africa Tech Workers Movement, who reviewed the situation with affected workers directly, concluded the termination was driven by a desire to distance Meta from people who had spoken publicly about what they saw. Meta has not addressed that characterisation.
The legal responses followed quickly. The UK's Information Commissioner's Office contacted Meta describing the Swedish reports as concerning and seeking clarification. Kenya's Office of the Data Protection Commissioner opened a formal inquiry. A class action lawsuit was filed. The Electronic Frontier Foundation issued an advisory directed at consumers of Ray-Ban glasses. These are not the reactions to a disputed or marginal story. They are the institutional responses to a credible, documented account of privacy practices that bypassed consent on a significant scale. Each institution that responded has its own threshold for what triggers a formal inquiry. All of them were triggered here.
What the Ray-Ban footage situation reveals about AI data pipelines deserves direct treatment rather than being handled as context. Meta sold seven million pairs of Ray-Ban glasses in 2025 alone. Each pair can record first-person video. Training a vision model to understand what those cameras see in the real world requires human annotation of real-world footage. Real-world footage collected from millions of people wearing smart glasses in their daily lives will inevitably include footage of bathrooms, bedrooms, intimate moments, and private environments, because those are places people go while wearing glasses. The annotation pipeline that processes that footage requires human reviewers. Those reviewers are overwhelmingly employed in low-cost offshore markets where workers have limited legal protections and where the relationship between the end client, Meta, and the direct employer, Sama, creates diffuse accountability. The pipeline is legal, widely used across the AI industry, and produces outcomes that bear no resemblance to the lifestyle product marketing Meta uses to sell the glasses to consumers.
The psychological dimension of annotation work is documented in academic research and largely absent from how the industry talks about itself publicly. Studies of content moderators and data annotators consistently show elevated rates of PTSD, anxiety disorders, and secondary trauma among workers who review graphic, violent, or sexual content at scale. Sama has marketed its mental health support offerings, including on-site counselling services, as evidence of responsible practice. On-site counselling is not a substitute for not requiring workers to review footage of non-consensual sexual activity as a job function. The distinction matters, and the workers who spoke to journalists made it clearly: the counselling was available, the content was still in the queues, and the combination was what they wanted to go on record about.
The product trajectory makes this more urgent, not less. Meta is now shipping prescription Ray-Ban glasses, targeting the roughly 2.2 billion people globally who already buy corrective lenses. Turning AI-equipped cameras into default vision correction devices is a distribution strategy that, if successful, puts first-person recording capability on billions of faces. The vision model that powers the AI features in those glasses needs to perform well across every context its users encounter, including private and sensitive ones. The data pipeline that trains it follows. Whether Meta redirects that pipeline to a different contractor after Sama, automates annotation through synthetic data generation, or changes the scope of what footage gets submitted for training is not something the company has disclosed. The 1,108 workers who lost their jobs are not privy to that decision either. They found out their employment was ending six days before it did.
The workers who spoke to the Swedish journalists did not create the problem they described. They described the problem that already existed, which is what whistleblowing is. The outcome for them is that the contract their employer depended on was terminated, their employment ended with statutory minimum severance, and a Meta spokesperson framed the whole thing as a vendor performance issue. The regulators who opened inquiries, the lawyers who filed the class action, and the journalists who reported the story did not produce that outcome. Meta's decision to frame contract termination as a standards enforcement action rather than a reputation management response produced it. The distinction is worth keeping clear.
","excerpt":"Over 1,100 Sama workers in Nairobi lost their jobs in April after Meta terminated its contract following worker disclosures that they had been assigned to review intimate footage captured without consent by Ray-Ban smart glasses.
Also read: He Jiankui is back in a lab and this time he is building brain-computer interfaces • Anthropic Built a Model Too Dangerous to Release and Boards Still Have No Framework for What That Means • Agentic AI is moving from boardroom buzzword to operational reality in 2026
