Plaintiffs in New Jersey and California are suing Meta for alleged privacy violations after anonymous contractors stepped forward to reveal intimate videos from the company’s AI smart glasses are being surreptitiously shared. The privacy lawsuit claims that Meta’s marketing tagline of “designed for privacy, controlled by you” is false as the glasses may take videos unbeknownst to their users and send them to contractors in Kenya for review.
The case reflects similar issues that have previously occurred with an assortment of smart home assistants and devices that would share random privacy-invasive samples for “quality control” and AI training purposes, but Meta claims that its AI smart glasses will not forward any video to contractors unless users choose to share it. Their claim is contradicted by anonymous whistleblower reports from amongst the ranks of the contractors reviewing this footage, which apparently have enough substance to have prompted a UK Information Commissioner’s Office (ICO) investigation.
Suit claims AI smart glasses independently shared “deeply private” video clips
Among other things the whistleblowing contractors note that the AI smart glasses share videos of people engaged in sexual activity, using the bathroom, caring for children, and reviewing financial and sensitive documents. The contractors say that if they raise concerns about viewing such material, they are let go from their jobs. They additionally say that these clips are supposed to be protected by a “face anonymization” technology that blurs out the identity of the people in the clip, but that it sometimes does not work.
The central claim of the privacy lawsuit is that Meta’s Ray-Ban AI smart glasses are “pervasively” advertised as protecting sensitive user data of this nature. It notes that in April 2025, Meta altered the privacy policy for the glasses to establish that they must always be on for the sake of certain AI features. Some of the video apparently comes from users that were unaware the glasses might still be recording when taken off, for example when set down on a bedside stand or a sink.
The subcontractor that processes the videos is identified in the privacy lawsuit as Sama, a company based in Nairobi. Sama employs thousands of people tasked primarily with data annotation, or manually labeling objects seen in videos (such as cars or furniture) to help train the AI smart glasses. The lawsuit asserts that this practice has been in place since 2021 and that Meta has been aware of the privacy risk, but has failed to adequately disclose the existence of this human review pipeline to consumers.
Claims in privacy lawsuit reflect prior incidents
Meta has responded to the story and the privacy lawsuit by claiming that videos are only passed on to Sama when the user asks a related question and explicitly allows the video to be uploaded. This would seem to fly in the face of claims of sexual activity, bathroom use and financial statements being reviewed being among the included videos, but thus far the only evidence is testimonial from anonymous whistleblowers.
The privacy policy for the AI smart glasses does not entirely clear things up. While it says that videos that users proactively record (such as private messages) are not transferred anywhere off-device without user permission, it is more unclear about the essentially always-on “multimodal” features such as Live AI. In this area the privacy statement says that “images” that are not present on the user’s camera roll may be used for “training purposes.” It does not elaborate on whether that means human review, though the UK version of the privacy policy very briefly mentions the possibility.
While this is a fairly new and novel issue involving AI smart glasses, problems with “quality control” clips being surreptitiously passed to third-party contractors have been an issue for almost a decade now. Most of the major voice assistants went through a wave of this from about 2018 onward, activating without their “wake word” (and unbeknownst to the device owner) and sending sometimes intimate clips to such contractors for training and review. A similar story unfolded in 2019 when an anonymous whistleblower revealed that Google Assistant was randomly capturing audio snippets for training without the “Hey Google” wake word being spoken, including similarly private and intimate moments. Amazon’s Alexa was also found to be doing something similar, sparking privacy lawsuits. Apple’s Siri also did this prior to iOS version 13, though the company has since disabled recording by default and now sends any captured audio to its own internal employees rather than contractors (after a similar whistleblower report).
The Ray-Ban glasses use a similar “Hey Meta” activation phrase for AI features, something that other assistants have struggled with in the past. Testing has found that some assistants are regularly triggered by thousands of word combinations that are somewhat similar to the activation phrase, as well as background audio such as television. According to their privacy policies, companies sometimes take this as permission to upload audio clips to their servers and potentially pass them on for review.
