For two years Meta has offered the same answer to the obvious question about camera glasses: a small white LED on the right rim tells everyone else when the camera is running. This week the company conceded, for the second time in under two months, that the answer had a hole in it — and that the hole was closed by software, not by the slogan.
The bypass was almost insultingly simple. Meta’s glasses already refuse to begin a recording if the capture LED is blocked. But nothing stopped a wearer from starting the recording first and covering the light a second later. Tutorials demonstrating the trick have circulated on Instagram and TikTok for months.
“The camera won’t start recording if the LED is blocked, but some people try to bypass this by starting a recording and then covering it up,” Alex Himel, Meta’s vice president of augmented reality, wrote in a post on Threads. “A fix for this is starting to roll out — the camera will now stop working if the light is covered during a recording.”
Himel framed the persistence of the workarounds as vindication rather than failure. “Only a tiny minority of people try to get around these protections, and the fact they have to go to these lengths shows our efforts are working,” he wrote.
Two patches and a billboard
The update is the second privacy patch in seven weeks. In July, Meta pushed a mandatory update disabling the camera entirely if the glasses detect the capture LED has been physically tampered with or destroyed — a response to a cottage industry of hardware mods. That fix addressed destruction of the light. It did not address a fingertip, a strip of tape, or the roughly two-dollar stickers Engadget reported were being sold specifically to defeat the indicator without triggering detection.
Alongside the patch, Meta is opening what a company spokesperson described as an education and awareness campaign explaining how the LED works. The first billboard is up in Los Angeles, showing the glasses with the recording light lit above the line that Meta Glasses are designed for everyone, not just the people wearing them. The linked privacy page also asserts that you, and only you, can see the photos and videos captured on the glasses — a claim, as 9to5Google noted, that sits uneasily beside earlier reporting that human moderators review some footage from Meta’s AI display glasses.
The volumes are why any of this matters. EssilorLuxottica said it sold more than 7 million Ray-Ban Meta AI glasses in 2025, more than triple the roughly 2 million sold across 2023 and 2024 combined, and Bloomberg reported in January that the partners were discussing raising annual capacity to 20 million units by the end of 2026. The $799 Ray-Ban Display, with a monocular in-lens screen and a neural wristband, has been supply-constrained since launch. TrendForce projects global AR glasses shipments of about 950,000 units in 2026, up 53 percent year on year, with Meta the main driver.
Analysis
An LED is a nineteenth-century solution to a twenty-first-century problem. It communicates one bit — recording, or not — to anyone who happens to be looking at the right corner of a stranger’s eyewear, understands what the glow means, and is close enough to see it. Every one of those conditions fails routinely in the situations bystanders actually worry about.
“Meta glasses do have a light indicator that turns on when recording is active, but people might not see it or know what it means,” said Adam Schwartz, privacy litigation director at the Electronic Frontier Foundation.
The deeper problem is that the LED signals capture, not inference. Always-on multimodal assistants do not need to save a file to extract value from a scene, and Himel’s phrasing is careful: the light alerts bystanders when media is captured for someone’s gallery. What an on-device model reads, transcribes or recognises in the moment has no indicator attached to it.
Facial recognition is where that gap becomes untenable. WIRED reported in June that Meta had quietly shipped a dormant system called NameTag inside the Meta AI app — downloaded more than 50 million times — designed to convert faces captured by the glasses into biometric signatures stored on the wearer’s phone. Meta stripped most of it within a day and said no final decision had been made. Meta also licensed facial recognition and liveness detection from ROC supporting up to 10 million face templates, and a patent published on 13 August, US 2026/0238876 A1, describes an assistant using facial and expression recognition to decide which people in a frame are interesting enough to keep in shot.
“Putting cameras into glasses is an inherent privacy threat, and putting face recognition on the camera and the glasses is an intolerable escalation of that privacy threat,” Schwartz said.
Aanjhan Ranganathan, an associate professor at Northeastern University’s Khoury College of Computer Sciences, called the code’s removal good but the trajectory revealing — the fact it reached so many users at all, he said, “tells you the default direction the product is heading absent external pressure.” His concern is what recognition does to public space: it changes what a stranger can learn about you “from essentially nothing to a full dossier.”
That is the structural asymmetry no patch resolves. The wearer consents by buying the product. The bystander consents by walking down the street.
What to watch
Three things. Whether the fix ships to all models or only current hardware, and how quickly the accessory market routes around it — the sticker sellers have beaten two updates already. Whether Pennsylvania’s proposed bill, which would mandate visual recording indicators and bar wearers from disabling them, moves; it is the first state proposal aimed at smart glasses specifically rather than biometrics generally, against a federal vacuum where no BIPA equivalent exists. And whether Meta decides on facial recognition before Google ships its own glasses with Warby Parker and Gentle Monster later this year. The ACLU and 75 organisations have already asked Meta not to. A billboard in Los Angeles is not an answer to that question.
“Putting cameras into glasses is an inherent privacy threat, and putting face recognition on the camera and the glasses is an intolerable escalation of that privacy threat.”— Adam Schwartz, Privacy Litigation Director, EFF