Meta's 'super sensing' smart glasses: always-on recording and the privacy questions it raises

Smart glasses have so far been a modest technology, capable of snapping the occasional photo or playing audio. A new report suggests Meta is aiming at something far more ambitious and far more contentious: eyewear that is always watching and listening. According to the Financial Times, as relayed by The Verge, the company is working on prototype super sensing glasses designed to be continuously aware of the world around the wearer.
The described capabilities are a leap beyond current devices. The prototypes could continuously record audio and snap photos every few seconds, building a running sensory record of wherever the wearer goes. That stream of data would feed Meta's AI, allowing the wearer to ask the assistant about things it had seen or heard, effectively giving the glasses a memory of the day.
The appeal of such a device is easy to imagine. An assistant that quietly registers everything could remind you where you left your keys, recall a name you were told an hour ago, or summarise a conversation you half-missed. For the wearer, the promise is a kind of perfect recall, an AI companion that never stops paying attention on your behalf.
But the same feature that makes the glasses useful to the wearer makes them fraught for everyone else. A person walking past, a colleague in a meeting or a stranger on a train has not agreed to be recorded, yet they could be captured continuously by a device that gives no clear signal it is doing so. The long-standing worry about camera glasses, that they erode the ordinary expectation of not being filmed, becomes far more acute when recording is constant rather than deliberate.
One detail in the report stands out. The FT describes a proposed system in which the raw footage and audio would not be stored by Meta or made available to the user; instead, only metadata derived from the audio and images would be kept. The apparent aim is to reassure people that the actual pictures and recordings are not being hoarded, only machine-generated descriptions of them.
That design choice is double-edged, and it deserves scrutiny rather than reassurance at face value. Not storing raw footage could genuinely reduce certain privacy risks, since there would be no library of images to leak or subpoena. But metadata can itself be deeply revealing, describing who was present, what was said and where, and a system that continuously generates such descriptions is still a continuous surveillance system by another name.
It is important to be clear about the status of these claims. The report describes prototypes and proposed systems, not a shipping product, and companies routinely explore designs that never reach consumers. Meta has not committed to releasing glasses that work exactly as described, and the details of any eventual device, including its privacy safeguards, could differ substantially from early plans.
The episode nonetheless illustrates a defining tension of the current AI wave. The most powerful assistants are the ones with the most context, and the richest context comes from sensors that see and hear as much as possible. That logic pushes relentlessly toward always-on capture, which collides directly with the privacy of everyone who happens to be nearby but never opted in.
Regulation and social norms will struggle to keep pace. Laws on recording vary widely between countries and even regions, and were mostly written for an era of deliberate, visible cameras rather than ambient, continuous ones. Whether a bystander has any say over being captured by a passing pair of glasses is a question existing rules answer inconsistently, if at all.
For now, the report is best read as a signal of intent and of direction. The technology industry is clearly betting that ambient, always-aware AI is the next major interface, and glasses are the natural vehicle for it. Whether society accepts a world where the people around you may be quietly recording at all times is a question no prototype can answer, and one that will be decided far outside any engineering lab.
Read next

Can video games teach robots to move? Inside the bet on a 'ChatGPT moment' for robotics
The startup General Intuition is betting that millions of hours of video game footage can train foundation models for physical AI, making smarter robots easier to build with far less real-world data. The idea points to why robotics has lagged behind language AI, and what a breakthrough might require.

Right to repair explained: what the John Deere FTC settlement means for owners
John Deere equipment owners will gain broader rights to repair their own machines under a settlement with the US Federal Trade Commission. The case is a milestone in the long-running right-to-repair fight, which asks who is really allowed to fix the increasingly software-laden products people buy.

What is the Fiat Topolino? America's cheapest new EV, and the micro-car idea behind it
The Fiat Topolino is arriving in the US at about $13,995, making it the country's cheapest new electric vehicle, but it tops out at 19mph and travels roughly 46 miles per charge. It is less a conventional car than a micromobility vehicle, and it poses a real question about what an affordable EV should be.

AI cheating in universities: what happened when one professor brought back the in-person exam
When an Ivy League professor grew suspicious that students were leaning on AI, he replaced take-home work with a supervised, in-person final. Scores fell by about half, a result that has become a flashpoint in the wider debate over what chatbots are doing to how students learn and how universities assess them.

What is a firmware backdoor? The Tenda router flaw explained
A security advisory reports that multiple versions of Tenda router firmware contain a hidden authentication backdoor, letting attackers bypass the login. Here is what a firmware backdoor is, why home routers are a favorite target, and how to protect your network.