Category: Technology | Published: 2026-05-19
When The Camera Never Really Switches Off
Meta's Ray-Ban smart glasses have attracted plenty of attention since they launched, mostly positive. They look like ordinary sunglasses, they stream video, they connect to AI, and they sit on your face with no obvious indication to the world around you that anything unusual is happening. That last part is precisely what makes the latest controversy so significant.
In February 2026, workers employed by data annotation firm Sama in Nairobi gave interviews to Swedish journalists describing what they had been reviewing as part of their job. The material, captured by users of Meta's AI-powered smart glasses, included footage of people undressing, using the toilet, and handling sensitive personal documents. According to one worker, the job amounted to seeing everything, from living rooms to intimate situations, with no filter and no warning about what might appear.
Within two months of that investigation being published, Meta had terminated its contract with Sama, citing a failure to meet standards. More than 1,100 Kenyan workers received redundancy notices with just days to prepare. The official explanation was thin, and the episode has since attracted regulatory attention on several continents.
What The Workers Were Actually Doing
To understand why this matters, it helps to understand what data annotation actually is. AI-powered smart glasses do not arrive ready to recognise faces, interpret scenes, or understand context. That capability is built by exposing AI models to enormous amounts of human-labelled visual data. Someone has to watch the footage, categorise what is happening, flag objects and actions, and confirm that the AI's interpretation is correct.
That work is real, human, and often carried out at significant scale in lower-wage markets, sometimes without the workers having full visibility of what they might encounter. In this case, the footage came directly from wearable cameras worn by real people going about their daily lives. The gap between what those users thought was happening to their footage and what was actually happening to it is the heart of the problem.
Meta has acknowledged that human review is part of its AI training process, and that photos and videos are used with user consent to improve product performance. The question that regulators, privacy groups, and labour campaigners are now pressing is whether users who consented to data being used for AI training genuinely understood that real people would be watching footage of their most private moments.
Why Meta Ended The Contract, And Why That Is Disputed
Meta's stated reason for terminating Sama was that the firm did not meet its standards. Sama has strongly denied this, stating publicly that it had consistently delivered against the operational, security, and quality standards it was contracted to meet, and that no concerns had been raised before the decision was made.
Naftali Wambalo of the Africa Tech Workers Movement put the challenge bluntly, suggesting that the standards being referred to were less about quality and more about the workers having spoken to journalists. Meta has not publicly responded to that characterisation. What is clear is that more than a thousand people lost their jobs with minimal notice at the same time that the story was generating significant international coverage.
This is not the first time Sama has been involved in a controversy of this kind. Earlier data moderation contracts with Meta, involving the review of harmful content, have previously been linked to accounts of psychological harm, inadequate support, and poor pay. Sama exited parts of that work some time ago, but the broader picture of how outsourced AI labour operates under these contracts is one that regulators are looking at with increasing seriousness.
The Regulatory Response
The fallout from the investigation has reached multiple jurisdictions. The UK's Information Commissioner's Office described the reports as concerning and asked Meta for further information. Kenya's data protection authority opened its own investigation. A class action lawsuit in the United States alleges that Meta misrepresented how the smart glasses handle user privacy.
In Europe, the questions around AI-powered smart glasses have been running for some time already. Wearable cameras that blend into ordinary eyewear create a specific challenge for data protection law, because bystanders who appear in footage have no obvious way of knowing they are being recorded, and the path that footage takes between capture, storage, transmission, and human review is not always visible to users themselves, let alone the people caught on camera.
What This Means For The Way AI Actually Works
The deeper point here is one that tends to get lost in discussions about how impressive modern AI has become. These systems are not purely automated. At multiple points in their development, and often in their ongoing operation, they depend on human beings to review, label, and validate the data they learn from.
For AI-powered smart glasses specifically, that means footage captured in everyday life, sometimes including sensitive situations, is passing through a pipeline that extends well beyond the device in your hand or on your face. The speed of AI development has often outpaced the transparency of how that pipeline operates, and the Sama controversy is one of the clearest examples yet of the gap that creates.
As AI-powered wearable devices become more capable and more common, that pipeline is only going to process more data, not less. Managing it responsibly is not optional, it is a legal and reputational necessity.
What Your Business Should Take From This
Most UK businesses are not building their own AI-powered smart glasses. But the lessons from this episode apply more broadly than you might expect.
First, if your business uses AI tools, whether in customer service, operations, or data analysis, it is worth understanding where the data those tools rely on actually comes from and how it is processed. Under UK GDPR, responsibility for personal data does not end when it is handed to a third-party provider. You remain accountable for what happens to it, including during AI training or model improvement processes that a vendor may carry out on your behalf.
Second, supply chain transparency matters. Meta's position, that it did not know Sama was falling short, does not provide any legal protection and creates obvious reputational risk. Knowing who your data partners are, what standards they operate to, and how those standards are verified is increasingly a basic expectation from regulators and customers alike.
Third, worker welfare is not a separate issue from data governance. When the human layer behind AI is poorly protected, the outputs of that work are less reliable, and the exposure for everyone upstream is greater.
This is the kind of governance thinking we help businesses apply through our cyber security services. Understanding your data lifecycle, including what happens when AI tools are involved, is part of running a secure and compliant organisation. The smart glasses story is a vivid reminder of how quickly gaps in that understanding can turn into headline-level problems.