The unsettling part of a new Meta smart-glasses patent is not the camera. Cameras are already on faces. It is the suggestion that the glasses could decide which people and moments are worth capturing before the wearer presses anything.
The published application describes an assistant system that can use head-mounted cameras to detect people, associate them with facial recognition, detect actions, and generate media files according to privacy settings. Privacy Guides surfaced the filing this week. The document is real; a product announcement it is not.
That difference matters. Patents are broad defensive and exploratory documents. They often describe techniques that never reach a device, and no reader should treat this one as proof that today’s Ray-Ban Meta glasses recognise everyone in view or secretly auto-record them.
Still, it exposes the design choice the category cannot postpone: when a wearable camera can understand a scene, who gets to set the rule for the people who did not choose to wear it?
What the patent actually describes
The filing’s central idea is a system that processes visual data from a head-mounted camera, identifies people and actions, and can create media around them. Its claims include checking a privacy setting associated with a person of interest: one person might be eligible for identification and recording in a media-sharing app while another is not.
| Patent concept | What it could enable | What it does not prove |
|---|---|---|
| Facial identification in camera footage | A wearer’s device could distinguish one known person from another | That current retail glasses perform this function |
| Action detection | Software could flag moments such as a person moving or interacting with an object | That Meta is collecting an action log of bystanders |
| Person-specific privacy settings | A system could refrain from making media for an opted-out person | That a workable, universal opt-out system exists |
| Automatic media generation | A device could assemble clips without a manual record button | That Meta intends to ship automatic recording |
The last two rows are the collision. Recognition and automatic capture sound convenient when the subject is a friend who wants a shared highlight reel. They feel fundamentally different when the subject is a stranger in a shop, on transit or at school.
Meta’s current safeguards are narrower—and concrete
Meta’s July privacy explainer says current AI glasses use a capture LED that blinks during photo or video capture and that there is no off switch. The company also says its second-generation glasses disable the camera if they detect that the LED is blocked; an update extends that response to physical tampering or destruction of the light.
Those are worthwhile hardware behaviours. They address an important social signal: can people nearby tell that a camera is recording?
But a capture light cannot communicate the entire software stack. It cannot tell a bystander whether the image is being analysed locally, whether it is being uploaded, whether a name is being attached, or whether a moment was selected by an assistant. As we argued when noRecognition made camera privacy wearable, the privacy question increasingly begins before a human has time to inspect a recording.
The impossible consent problem
The patent’s use of person-level settings hints at an answer: recognise someone, then honour their preference. In a controlled group—family members sharing media, colleagues on a project—that may be workable. In public, it presents an awkward bootstrap problem. A system must identify or match a person before it can learn that person asked not to be identified.
That is not merely a technical wrinkle. A voluntary opt-out list can create a new biometric database, put the burden on bystanders to join it, and leave everyone else as the default raw material for a wearable camera.
The better consumer test is simpler. Smart glasses should make their recording state obvious, minimise on-device analysis where it is not essential, and avoid turning strangers into named, searchable inputs by default. If a genuinely useful feature requires recognition, the company should explain what is processed, where it goes, how long it is kept and how a non-user can object.
Bottom line
Meta’s patent is not a leaked feature list. It is a glimpse of the capability space engineers are considering, and that is enough to merit scrutiny now. The capture LED is a good start for visible recording. The next privacy challenge is less visible: whether glasses gain the power to decide who matters in a scene before the wearer does. That decision needs stronger rules than a blinking light.