About 7,000 Meta AI glasses just lost their cameras. A tampered privacy light is why.

About 7,000 pairs of Meta's AI smart glasses, the ones whose owners drilled out, destroyed, or otherwise tampered with the small recording-indicator light on the frame, just had their cameras permanently killed by a remote software update. It is the first time Meta has used a server-side action to break a feature on its own shipped hardware in defense of a privacy indicator rather than a defect.

The number is small. Fewer than one tenth of one percent of the 7 million units Meta sold in 2025, according to Business Insider via iHeart. But the mechanism is new. When the glasses detect that the capture LED has been physically tampered with, not just covered with tape or a sticker, the camera shuts off and stays off, even if the user later repairs the light. Earlier safeguards only blocked recording when the LED was already obscured at the start of a clip. The new system closes the loophole where users could begin recording and then cover the indicator. The enforcement, per Meta's AI Glasses FAQ, is server-side, which means there is no physical kill switch on the device. The software update from Meta is the kill switch.

Wearable cameras have never had a privacy signal that survived a determined user. Phones do not flash when they record video, and most laptop webcams can be taped over indefinitely. The market has settled on covering the light as a routine workaround. Meta's CTO Andrew Bosworth made the case in unusually direct terms at a recent press event: "You haven't had a camera that put a light on when it took a photo or video of somebody since you had like a VHS tape recorder in the 90s... Your phone doesn't have that, and it's the most commonly held camera on earth." The argument: a recording light is the kind of privacy norm cameras should have had all along, and Meta is now willing to break its own hardware to defend it.

The enforcement is also aimed at the small but visible market that grew up around the bypass. Meta is removing ads and Facebook Marketplace listings that sell tampering services, warning that accounts offering such modifications may be banned, and signaling possible legal action against businesses that offer LED-disable work, even outside Meta's platforms. The company is treating a hardware-mod aftermarket the way it has treated counterfeit goods or phishing services: as something to delist, de-platform, and litigate out of existence. The affected pairs are the proof that the market was real.

The ACLU and dozens of co-signing organizations have called on Meta to abandon plans to add facial recognition to its smart glasses, a feature the LED indicator would not constrain at all. The light tells bystanders that recording is happening. It does not say what the on-device AI is doing with their face. Bosworth's "your phone doesn't have that" line invites the obvious counter: phones are not worn on someone's face, pointed at strangers from across a room. The enforcement tightens the recording-signal norm. It does not answer the surveillance question.

The remaining question is whether a privacy indicator can be enforced at all without breaking the device. Meta's answer is yes, if you control the software. Privacy advocates' answer is that an indicator only signals recording, and the harder problem, what on-device AI does with bystanders' faces, still has no on-frame solution. Both are right, and neither closes the gap.