Who Saw the Bathroom Footage from Meta's Glasses?

The most ordinary scene is the worst one. A California plaintiff, identified as PL18, says he set the glasses on a bathroom counter and later noticed photos of family members using the bathroom and bathing in his gallery. Footage no one in the house meant to take. Steven Beltran, another California resident, alleges the gadget "falsely activate[d] twice a day." A Los Angeles wearer, PL01, used Meta AI during an "intimate moment" in his bedroom, then kept using the same frames to play music and ask questions. The complaint says every activation of the AI features, intentional or accidental, captures video and audio, ships it to Meta's servers, routes it overseas, and puts it in front of workers who watch, label, and embed the moments into models.

The reporters got there first

This is not a rumor that began in a filing mill. In February, Svenska Dagbladet and Göteborgs-Posten published a joint investigation after interviewing more than thirty people at Meta's subcontractor Sama and former Meta employees. Workers in Nairobi who reviewed glasses footage told the papers that faces sometimes appeared even when they were supposed to be blurred. Bathroom visits and undressing showed up on their screens. Shop staff in Stockholm and Gothenburg had told reporters the wearer controlled what was shared. The journalists chose "no" on the extra-data prompt in the Meta AI app and still had to send video through Meta's infrastructure to use the assistant. There is no local seeing.

After the investigation, Meta paused and then ended its Sama contract. A spokesperson told the Swedish papers the company takes the claims seriously, that photos and videos are private to users, that humans review AI content "to improve product performance, for which we get clear user consent," and that Sama did not meet Meta's standards. Sama announced a redundancy process affecting 1,108 workers. Naftali Wambalo of the Africa Tech Workers Movement told the BBC he believed the contract died because staff had spoken. Meta kept the product.

A name tag in the app

A separate suit, filed in Illinois in early September, alleges Meta pulled biometric identifiers from Facebook and Instagram photos to train NameTag, an unreleased face-recognition system for the glasses, plus image models including Emu and Muse Image. WIRED found NameTag code in the Meta AI companion app in June, an app downloaded more than fifty million times, then watched Meta strip the code the next day. Andrew Bosworth, Meta's chief technology officer, called the reporting dishonest, then described on a podcast a feature that would let a wearer remember names of people previously met. The company says it is not building a universal face database and that nothing has shipped. The ACLU and more than seventy organizations asked Meta in April not to release facial recognition on the glasses.

Meta's public answer to the Times is the industry shrug. "If you use Meta AI, we may review that data to help improve our products and people's experiences," a spokesperson said. "This works the same way as many other companies. We take steps to filter this data to help remove identifying information and to protect people's privacy." A recording light is not a filter. Apple now signs iPhone photos at the moment of capture, so later software can be checked against the original. Meta's glasses send the original into a training queue and call the LED consent.

Connect opens Wednesday in Menlo Park. Mark Zuckerberg will sell the next computing platform from a pair of frames. The complaint asks a duller question that will still be true when the keynote ends. Who saw the bathroom. The wearer thought the audience was the assistant. The assistant is a routing system with a human at the far end, paid to teach a model what a toilet looks like. Ending the Sama contract punished the people who described the work. It did not retire the work. Until the camera stays off unless a person who can be named has asked for the file, the LED is a prop.