Product1 publisher3 min readPublished
More than 70 Meta glasses plaintiffs say their home footage reached reviewers in Kenya
An amended class action in Northern California alleges every Meta AI activation on the glasses routed video overseas to low-paid labelers. Some of the plaintiffs were never users, only people who got recorded. Meta intends to fight.
The Product Desk · Product desk
What happened
- More than 70 people who bought, used or were recorded with Meta's smart glasses have sued the company, saying their intimate images reached paid workers abroad who review and label footage.
- The complaint runs 230 pages in a Northern California federal court and accuses Meta of fraud, false advertising and violations of consumer protection laws in various states.
- Meta says it may review Meta AI data to improve its products and takes steps to filter identifying information out, and the company intends to fight the allegations.
Compiled by The Product DeskSomething wrong?How this is made
Why it matters
- exposure Household members and bystanders are inside this plaintiff group, so the consent question for a camera product reaches people who never accepted terms and cannot be reached by changing them.
- contradiction Meta's published description of blurring and the reviewers' account of what they saw cannot both describe production. A buyer choosing this device is left to guess which one does.
- decision Teams that pay an outside vendor to label captured media now have to settle at contract time whether raw frames may reach a human reviewer. What the complaint goes after is the handling itself.
- precedent With Disney and Apple already facing biometric collection suits, plaintiffs' firms have a template for arguing that consent fails in the data handling chain itself.
According to the complaint, a California user it calls PL18 put his smart glasses on the bathroom counter and they recorded without his knowledge [4]. The lawsuit says "photos of his family members using the bathroom and bathing began appearing in his gallery" [5].
Up to that point this is a hardware complaint about a camera that starts when it should not. The rest of the filing is about where a picture goes once it exists. The complaint says: "Every activation of the AI features, intentional or accidental, captures video and audio that is transmitted to Meta's servers, routed overseas, and reviewed by low-paid human workers who watch, label, and embed these moments into Meta's AI models" [7]. The reviewers are in Kenya, working for companies that help train Meta's AI, and the amended version of the proposed class action was filed in late August [2].
Meta's answer describes the same route. "If you use Meta AI, we may review that data to help improve our products and people's experiences - this works the same way as many other companies," a Meta spokesperson said [10]. The company says it takes steps to filter identifying information out of that data, disagrees with the allegations and intends to fight them [10][11].
So both accounts have someone outside the household looking at the footage, and they differ on what that person sees. Meta's website says that when a user asks Meta AI about what is in front of them the camera activates, that faces and license plates are blurred, and that only the people who take photos and videos can see them unless they share [9]. Some of the Kenyan workers who reviewed glasses footage told Svenska Dagbladet and Goteborgs-Posten that faces sometimes appeared even when they were supposed to be blurred; the two Swedish papers interviewed more than 30 employees at Meta subcontractor Sama and former Meta staff for the February article the lawsuit cites [13][12].
What users do is leave the glasses on the counter and say "Hey Meta" without thinking about a queue [8]. The plaintiff group includes people who were recorded as well as people who bought or used the device, which means part of this class never saw a settings screen at all [1][19]. Tina Wolfson, a principal at Ahdoot & Wolfson in Burbank and one of the lawyers for the plaintiffs, said: "People who bought these glasses and thought that they were cool gadgets weren't aware that every time they activated AI, video was sent to train Meta's AI" [14].
The privacy control Meta points to is a light that signals when the glasses are recording [16]. That light informs the person standing in front of the lens at the moment of capture, and it does not travel with the file into a labeler's work queue.
Before launch, anyone shipping a capture product already knows whether raw capture leaves their systems and reaches a human who is not their employee, and whether that capture routinely includes people who never installed anything. One of those on its own is a vendor management problem. Both together is the position Meta is now defending in a 230-page complaint alleging fraud, false advertising and state consumer protection violations [6], and the documents that decide it are the annotation contract and whatever redaction runs before a clip reaches a reviewer.
What to watch
- Whether the Northern California court certifies the proposed class, and whether the recorded-but-never-purchased plaintiffs survive that stage.
- Whether Meta revises what its site says about who can see Meta AI camera captures, or adds a redaction step before footage reaches a reviewer.
- Whether Sama or another annotation vendor is named as a defendant in a later amended complaint.