You probably know someone — a coworker at a SoMa startup, a Caltrain commuter, a guy at the climbing gym in the Mission — who's been rocking a pair of Ray-Ban Meta smart glasses lately. They look like any other pair of sunglasses. But a new investigative report published last week by two Swedish newspapers raises a question worth sitting with: that neighbor asking "Hey Meta" to help identify a plant in their apartment could, without knowing it, be streaming footage of a partner's bank card, a bedroom, or something considerably more explicit to a team of human contractors halfway around the world. And they'd have no idea.
The joint investigation, published February 27 by Svenska Dagbladet and Göteborgs-Posten, found that footage captured through the glasses — when users invoke the AI assistant by saying "Hey Meta" — gets routed to a Nairobi, Kenya data annotation firm called Sama, where human contractors review and label the video to help train Meta's AI systems. That, in itself, is a fairly standard AI industry practice. What isn't standard is what those workers say they're seeing.
Workers Describe an Eyeful of Other People's Most Private Moments
The contractors, who spoke anonymously out of fear of losing their jobs, described a parade of unintentional oversharing, according to the Swedish papers. Workers reported being asked to review highly sensitive and intimate data, including footage of people going to the toilet or getting undressed. One worker recounted watching a man set his glasses on his nightstand and leave the room, only for his partner to walk in and change clothes. Others described footage of people in bathrooms, sex scenes, and users watching pornography while wearing the glasses.
"We see everything – from living rooms to naked bodies. Meta has that type of content in its databases," one annotator told the Swedish outlets. Workers also described reviewing transcriptions of AI conversations covering everything from personal secrets to explicit sexual commentary. The consensus: the people on camera almost certainly had no idea what was being recorded. "I don't think they know," one contractor said, "because if they knew they wouldn't be recording."
That consent gap is compounded by a chilling workplace dynamic. Contractors described feeling pressured to process disturbing content without questioning its ethical implications. "You are not supposed to question it. If you start asking questions, you are gone," one worker explained. Many of these employees are college-educated young adults in Nairobi who, according to the report, fear losing their livelihoods if they speak up or push back.
The Privacy Fine Print Nobody Reads
Here's where it gets a little uncomfortable for Bay Area residents who might own a pair. Meta's AI Terms of Service — the ones you agree to by activating the glasses' AI features — do technically disclose that human review of AI interactions is possible. The terms state that "in some cases, Meta will review your interactions with AIs, including the content of your conversations with or messages to AIs, and this review can be automated or manual (human)." The document also warns users not to share "information about sensitive topics" with the AI — which is genuinely ironic, given that the sensitive information being captured appears to come from people who didn't realize they were sharing anything at all.
When Swedish reporters tested the glasses themselves and tried to avoid sending data to Meta by disabling the internet connection, the AI functions stopped working entirely — the glasses require a live connection to Meta's servers to operate. When users ask the AI assistant to identify a plant or read a sign, the glasses capture an image and transmit it to Meta's cloud infrastructure for analysis. There's no purely local mode. So much for staying in control of your own privacy.
Meta, for its part, took two months to respond to the Swedish reporters' repeated interview requests, then issued a statement referring them to its privacy policy without directly answering questions about what content filters exist, how long footage is stored, or whether video from users in countries like the U.S. ends up with contractors like Sama. A Meta spokesperson told Business Standard that data "is first filtered to protect people's privacy" and that the company "takes the protection of people's data very seriously." Given what the Kenyan workers describe, the filtering doesn't appear to be working particularly well.
Sama's Troubled Track Record
Sama, the subcontractor at the center of this story, is actually a California-headquartered company — Samasource Impact Sourcing, Inc. — with operations in Nairobi. It's also not a stranger to controversy. Back in 2021, Sama labeled tens of thousands of text passages containing depictions of sexual abuse, violence, and hate speech on behalf of OpenAI. According to a TIME investigation, Kenyan workers were paid roughly $1.32 to $2 per hour at the time. One worker described the experience as "torture." After further reports exposed worker trauma and alleged union-busting, Sama ended its content moderation work for Meta in 2023 — and shifted its focus to exactly the type of computer vision data annotation it's now doing with the AI glasses.
The company has also faced legal action from workers in Kenya. Four people in Kenya filed a petition calling on the government to investigate conditions for contractors reviewing content used to train large language models, alleging the work was exploitative and left some former contractors traumatized. Sama disputes the characterization of its working conditions, and has promoted itself as a mission-driven employer creating dignified jobs in low-income communities.
This Isn't Theoretical: Bay Area Already Saw This Coming
For locals, this isn't entirely abstract. In October 2025, SFist reported that the University of San Francisco issued a campus-wide safety advisory after multiple women reported being approached by a man wearing Ray-Ban Meta glasses. Officials described his behavior as "suspicious," and they believed he was posting the recorded interactions on social media under the handle @pickuplines.pov. The university said it was unable to identify all the students who may have appeared in the videos.
That incident laid bare what privacy advocates had long argued: the glasses' tiny white recording LED — the sole bystander notification that the camera is active — is nearly invisible in daylight, easily covered, and meaningless to anyone who doesn't already know what it signifies. Privacy-conscious residents have even started using a new smartphone app called "Nearby Glasses" that scans for the specific Bluetooth signals emitted by the frames to alert them when a pair is active. Meta declined to comment on the USF incident at the time, pointing instead to its terms of service advising users to "act responsibly."
And Then There's the Facial Recognition Question
Just when you thought the privacy picture couldn't get more complicated: TechCrunch reported in mid-February that Meta is actively planning to add facial recognition to the glasses through a feature internally called "Name Tag," which would allow wearers to identify people they see in real life using Meta's AI. The New York Times obtained internal documents showing Meta had specifically timed the rollout for what it called a "dynamic political environment" — in other words, when privacy and civil liberties organizations are already overwhelmed fighting other battles.
Meta previously shut down Facebook's original facial recognition system in 2021, citing "growing societal concerns," and faced years of legal scrutiny including a $5 billion FTC fine over privacy violations and biometric data settlements totaling more than $2 billion in Illinois and Texas. The company is now apparently betting the moment is right to try again — this time on 7 million pairs of glasses walking around in public.
What This Means Legally
California residents may have more protection here than users elsewhere, though the picture is murky. The California Consumer Privacy Act (CCPA) gives residents rights over their personal data and requires companies to disclose how they collect and use it. In February 2026, the Electronic Privacy Information Center (EPIC) specifically petitioned the California Privacy Protection Agency (CPPA) to investigate Meta’s glasses under the state's biometric information protections. But enforcement is uneven, and Meta's disclosures — buried in AI terms of service documents that the company itself warns users probably won't read in full — may or may not constitute adequate notice under California law.
Kleanthi Sardeli, a data protection lawyer at the European privacy watchdog None of Your Business (NOYB), told Svenska Dagbladet that users often don't realize the glasses are recording when they activate the AI assistant — creating a transparency problem with real legal teeth under GDPR in Europe. "Once the material has been fed into the models," she warned, "the user in practice loses control over how it is used." That principle applies whether you're in Munich or the Mission District.
The ACLU has also weighed in on the facial recognition piece. As The Verge noted, the organization's Nathan Freed Wessler called the technology "a uniquely dire threat to the practical anonymity we all rely on." Given that a U.S. Customs and Border Protection agent was reportedly photographed wearing Meta glasses at an immigration raid last year, the question of what happens when government agencies get interested in that data is no longer entirely hypothetical.
The Bigger Picture: The Hidden Human Cost of "AI"
The Swedish investigation also lands as a broader indictment of how Big Tech describes its AI products. When Meta markets the glasses as a sleek AI assistant that processes what you see, the word "AI" conjures algorithms and servers — not a room in Nairobi where someone is watching a video of your partner stepping out of the shower. The use of offshore data annotators in countries like Kenya, Colombia, and India to train AI models is industry-wide, and workers in those roles have had to review gruesome crime scene images and other disturbing content — a hidden human cost that companies tend to bury in lengthy terms of service.
As Help Net Security noted, Meta made AI camera and voice use the default setting in an April 2025 policy update, effectively burying the previously clear opt-out for voice recording storage. If you own a pair of these glasses and haven't dug into your sub-menus lately, it may be worth checking — because the defaults are not working in your favor.
