On a summer afternoon in August 2026, families at the Frog Pond on Boston Common were doing what families do at a spray pool. Children ran through the water. Parents held up towels so that small bodies could be changed out of wet swimwear. And according to witnesses who later spoke to police, a man in sunglasses was watching them through a camera.
The witnesses said he had been recording children on the playground, both with a phone and with a pair of camera-equipped sunglasses. Some said he followed a family who had been changing their children behind towels, scanning the children up and down with the glasses. When people confronted him, he took the glasses off and tried to snap them with his hands. Police seized the glasses and the phone. The man, a 60-year-old from Brockton, is a registered sex offender. In September he was arraigned on 69 further charges relating to photographing the intimate parts of children, and he pleaded not guilty. Nothing has been proved in court.
What makes the case so unsettling is not that a predator found a new tool. Predators have always found new tools. It is that the tool looked like nothing at all. A camera held up to a changing child is an obvious threat that any parent can see and act on. A pair of sunglasses is not. The incident, as Joy Buolamwini wrote in Nature in October, illustrates how such devices can subject bystanders to “intrusive surveillance without their knowledge or consent.” On this occasion, other people at the pool saw what was happening and stepped in.
That is the problem in miniature. Most of us move through public space on an unspoken assumption: we can see who is looking at us and roughly how. Smart glasses quietly break that assumption, and they are doing it at a scale, and with a degree of artificial intelligence, that the people who wrote our privacy laws never imagined. On 13 October 2026, Europe's data protection regulators finally put on paper what that means for the people on the other side of the lens.
Eight Pages From Brussels
The European Data Protection Board (EDPB), the body that brings together the European Union's national data protection authorities, adopted Statement 1/2026 on smart glasses at its October plenary. It is short: eight pages, 36 numbered paragraphs, signed by the board's chair, Anu Talus. It is described as preliminary. It is also, read carefully, one of the bluntest documents a European regulator has produced about a consumer product category.
Start with the controller question, which is the legal way of asking who is responsible. The board accepts that a wearer filming a friend for a private family chat may fall under the GDPR's household exemption. But it insists that exemption “must, however, be narrowly construed,” and that it vanishes when footage goes to a public social media account or the glasses are used for professional or political purposes. More importantly, it states that the GDPR still applies to the companies that supply the means of processing, even when the wearer is exempt. When a manufacturer or AI provider analyses an image from the glasses' camera “in which other persons appear,” it can be a controller in its own right. And controllers, the statement says, “should not rely on the users' actions to meet their own GDPR obligations, such as transparency toward other individuals whose data might be processed.”
The press release put it more plainly: neither the producer nor the provider of a linked service “can transfer their obligations to users.”
Then comes transparency, and here the board is almost confessional. “The current state of the art with regards to indicators on smart glasses leave many non-users unaware that their data is being processed,” it says. Privacy notices, the standard legal fix, “do not seem sufficient or appropriate to inform non-users,” because a manufacturer has no relationship with the stranger on the pavement and no way to contact them.
Consent fares no better. Where a manufacturer or AI provider processes data about non-users with whom it has no relationship, the board says it “would be difficult to obtain their freely given, specific, informed and unambiguous consent.” It closes an obvious loophole too: “Consent obtained after collection, for example following the recording, would not be valid.” You cannot film first and ask later.
That leaves legitimate interest, the flexible legal basis on which much of the data economy runs. The board does not rule it out. It does say that, given “the current insufficient transparency,” the balancing test “would be particularly difficult to demonstrate for further processing for commercial purposes that go beyond providing the service explicitly requested by the end-user, or for AI training, which bystanders wouldn't expect, especially if they are not aware of the initial data collection.”
Read that sentence again and notice what it targets. Not the snapshot of a sunset. Not the translation of a menu. It targets the business model: the reuse of what the glasses see to improve products and train models.
And then, in paragraph 8, the board steps outside the usual vocabulary of data protection altogether. Even when a device can be identified as smart glasses, it says, people may object simply because they are uncomfortable with the possibility of being recorded. “This could also result in a subtle social effect, where individuals alter their behaviour, suppress spontaneity, and conform to social norms perceived to be more socially acceptable due to their perceived threat of unvetted surveillance.”
That is not a sentence about data. It is a sentence about what public life feels like.
Who Notices the Little White Light
The industry's answer to the bystander problem has, for five years, been a light. Ray-Ban Meta glasses carry a small LED on the front of the frame that glows when the camera is capturing. Meta designed the glasses so that they stop recording if the light is covered with tape. The EDPB now wants considerably more: indicators “visible in many lighting conditions,” audio cues such as a shutter click, voice activation so that commands are spoken aloud, opt-in smartphone alerts, and frame designs that make smart glasses easier to tell apart from ordinary ones. It says these measures should be “impossible or difficult to disable (taking into account the state of the art)” and adds, pointedly, that “it should not be possible to record information when the indicator is covered.”
The footnotes are where the board's scepticism shows. Producers, it notes, should consider that LEDs may be invisible to people with visual impairments, and “that a white LED generally could be mistaken for a reflection or solar glare.” On a bright day, in other words, the main safeguard can look exactly like sunshine.
The light can also simply be removed. In October 2025, 404 Media reported that a hobbyist named Bong Kim was charging $60 to modify Ray-Ban Meta glasses so that the recording LED no longer lit up, while the glasses otherwise worked normally. The modification got around the tape-detection safeguard by changing the hardware rather than covering the light. Meta has since said it is cracking down on tampering services, and in July 2026 The Next Web reported that the company was patching its glasses to disable the camera if the light was tampered with. When Paris prosecutors announced in September 2026 that they had opened several investigations into women being filmed on the street without consent, Agence France-Presse reported Meta's response: that it had taken multiple steps to prevent tampering with the light.
Even an untampered light runs into a deeper limit, one that no firmware update can fix. A light only works if the person who sees it knows what it means. Researchers at Tsinghua University put that to the test.
Wearers and Bystanders Want Different Things
Their study was presented at CHI 2026, the leading academic conference on human-computer interaction, held in Barcelona in April. Its title is “Mind the Gap: Mapping Wearer-Bystander Privacy Tensions and Context-Adaptive Pathways for Camera Glasses,” and its authors are Xueyang Wang, Xin Yi and Hewu Li of Tsinghua University, with Kewen Peng of the University of Utah.
The first part of the research was a survey of 525 people in China: 232 current or potential wearers and 293 people who might find themselves on the other side of the lens. Each participant was given a short primer on how the devices work, including where the LED sits, and then shown six everyday scenarios. A travel vlog on a street. Shopping in a mall. A meeting with colleagues. A hospital consultation. A private party. A workout at a gym.
The pattern is stark. Wearers rated recording as reasonable in every scenario, including the gym among strangers. Bystanders reported significantly more privacy concern than wearers overall. When the researchers asked what bystanders expected and what wearers were willing to give, two gaps were statistically significant: bystanders wanted tighter control over how footage is shared, and they expected prior consent far more than wearers were prepared to seek it. The gym produced the widest gaps of all, and the hospital was not far behind. As the authors put it, the disparity grows in sensitive contexts “where bystanders' vulnerability peaks while wearers' willingness plateaus.”
Then there is what bystanders said they would actually do. Across the scenarios, between 65 and 90 per cent said they would take some form of defensive action. Camera avoidance was the most common response, chosen by between 52 and 80 per cent, peaking at 80 per cent in the mall. Nonverbal protest came next, at 31 to 51 per cent. Direct intervention, asking the wearer to stop recording or delete footage, was chosen by roughly a quarter to a half of participants depending on the setting. Passive acceptance was rare, at 10 to 15 per cent, except in the workplace meeting, where it rose to 35 per cent.
The LED did not reassure anyone. Only 41.3 per cent of bystanders and 35.3 per cent of wearers thought it was sufficient, and just 6.1 per cent of bystanders believed an LED alone was enough. Participants said it was too small, invisible in bright light, meaningless to people unfamiliar with the devices, and easy to block. One researcher interviewed for the study summed it up: the light “deters the honest but not the malicious.”
The finding that matters most for the burden question is about familiarity. You might assume the gap is just novelty, and that bystanders will relax once smart glasses become as normal as phones. The data says otherwise. Bystanders' concern stayed high at every level of familiarity, while wearers grew less concerned the more they used the devices. Experience widened the gap rather than closing it. The authors conclude that the gaps “reflect structural incompatibilities rather than inadequate goodwill.”
The study was conducted in China, and the authors are careful to say privacy norms differ between cultures. But the structure of the finding is not culturally specific. The person wearing a camera and the person in front of it are in different positions, and those positions produce different interests. Etiquette campaigns will not bridge that.
How Private Glasses Feed a Public Market
None of this would matter much if smart glasses were a niche toy. They are not. Buolamwini's Nature piece, published on 6 October 2026, reports that more than eight million smart glasses were shipped in 2025 and that by June 2026, sales of AI-equipped models had “reportedly risen by more than 700%” on the previous year. The first figure is well supported: the analyst firm Omdia counted 8.7 million AI glasses shipped globally in 2025, a 322 per cent increase, with Meta taking 85.2 per cent of the market. The second figure is harder to trace. Nature gives no primary source, and the growth figures published by market trackers vary widely with what they measure. Counterpoint Research, for instance, reported that global AI glasses shipments rose 263 per cent year on year in the first half of 2026. Whichever number you prefer, the curve is steep, and Omdia expects shipments to pass 15 million in 2026.
Scale matters because of what sits behind the lens. In April 2025, Meta told Ray-Ban Meta owners that AI features with camera use would be enabled by default unless they switched off the “Hey Meta” wake word, and that voice recordings would be stored to improve its products without an option to opt out, according to reporting by TechCrunch and others. Transcripts and recordings could be kept for up to a year, and owners who did not want their voice used for training had to delete recordings one by one. Those changes concern the wearer's data. But a camera pointed at a street does not capture only the wearer.
Then there is identification. In 2024, two Harvard students, AnhPhu Nguyen and Caine Ardayfio, built a demonstration system called I-XRAY. It streamed video from Ray-Ban Meta glasses to Instagram, used software to detect faces, ran them through the face search engine PimEyes, and pulled names, addresses and other details from people-search sites. The students said they would not release the code; the point was to show how easily off-the-shelf tools combine. In February 2026, The New York Times reported that Meta was planning a facial recognition feature for its glasses, known internally as “Name Tag,” which would let wearers identify people through Meta's AI assistant. According to the Times, an internal memo suggested launching “during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns.” Meta's plans, the reports noted, could change.
The EDPB saw this coming. Pairing smart glasses with facial or voice recognition, it wrote, “raises unprecedented questions regarding societal choices and poses a high risk to the rights and freedoms in public spaces.”
Put these pieces together and the idea of ordinary eyewear that can silently record, analyse and identify every person in the wearer's field of view stops sounding like hyperbole. Not every pair does all three today. The direction of travel is clear, though, and the hardware already in circulation is what tomorrow's software will run on.
What Happens to a Street That Might Be Recording
Which brings us to the first big question. What does all this do to the texture of public life when ordinary eyewear can record, analyse and potentially identify everyone in view, and nobody can tell whether it is doing so?
The sociologist Erving Goffman gave us the vocabulary in 1963. In his study of how strangers conduct themselves in public places, he described “civil inattention,” the small ritual by which strangers acknowledge each other with a glance and then look away. It is how a crowded street stays bearable. I see you; I am not going to examine you. The urbanist Jane Jacobs, writing two years earlier in The Death and Life of Great American Cities, made a related point about “eyes on the street”: the informal, mutual watching of neighbours and passers-by that keeps a city safe. Both describe a kind of looking that is reciprocal, fleeting and, crucially, visible. You can see the eyes that are on the street. They are attached to people you could look back at.
Camera glasses change both halves of that bargain. The glance that used to end in looking away can now be stored, replayed, shared and searched. And the eyes on the street are no longer necessarily attached to a person who is present. They might belong to an AI model on a server, a TikTok audience, or a face-search engine. Civil inattention depends on the other person choosing not to keep what they see. A device that keeps everything by default removes that choice from the encounter entirely.
We already know what this does to behaviour, because we have seen it online. In a 2016 study in the Berkeley Technology Law Journal, Jon Penney found a statistically significant drop in traffic to Wikipedia articles on privacy-sensitive topics after the 2013 revelations about NSA surveillance. The articles themselves had not changed. People's sense of being watched had. That is the chilling effect: not punishment, but the quiet edit people make to their own behaviour when they suspect an audience.
The EDPB's paragraph 8 is the physical-world version of that finding, and the evidence that it is already happening is accumulating. In August 2026, researchers at the University of Sydney published early findings from an analysis of 350 publicly available Instagram videos, posted between 2023 and 2026 and linked to pick-up artist communities. In a subset of covert point-of-view videos, around 60 per cent of interactions were classified as potential harassment, and in about 43 per cent of the videos women were exposed to abusive comments, some of which revealed personal details. The researchers found covert filming in workplaces, gyms and beaches. “Smart glasses have given pickup artists a new tool to further exploit women,” said Milica Stilinovic, a postdoctoral research associate on the project.
The BBC reported in 2026 on a woman in London who discovered that a stranger who had chatted to her near a shopping centre had filmed the whole encounter through smart glasses and posted it online, where it drew tens of thousands of views. In Paris, prosecutors described a “growing trend on social media involving women being filmed on the street.”
The CHI survey shows what the defensive response looks like in aggregate: most bystanders say they would avoid the camera, a third to a half would protest without words, and a substantial share would confront the wearer. Those are the behaviours of people who no longer trust the encounter. And the EDPB's list of sensitive settings, from medical facilities to swimming pools, is a list of exactly the places where people most need to be unguarded. Public space does not empty. It flattens. People become slightly more guarded, slightly more performative, slightly less willing to do the unplanned thing. That is what “suppress spontaneity” means in practice.
The cruellest part is that the effect does not require anyone to be recording. It only requires that recording be possible and impossible to detect. A street where one person in a hundred wears camera glasses is, from the bystander's point of view, a street where anyone might be. The chilling effect is a product of uncertainty, not of footage. You cannot opt out of uncertainty by checking for a light that may not be there.
There is a fairness issue, too. Smart glasses have genuine benefits, and the EDPB lists some of them, including audio assistance and hazard awareness for people with disabilities. Hamburg's proposed restrictions on wearing them in pools and public transport have reportedly included exemptions for people who rely on them for medical or accessibility reasons. Those uses deserve protection. But the benefits go to the wearer, while the costs of uncertainty fall on everybody else in the room. That asymmetry is the root of the burden problem.
Why Neither Wearer Nor Bystander Can Carry It
The second big question starts from a premise: that the burden of informing bystanders cannot realistically fall on either the wearer or the bystander. Is that premise right? The evidence says yes, for different reasons in each case.
Start with the wearer. The EDPB does say that users “should always ensure that they respect the privacy and dignity of other individuals,” and that explicitly telling people you are using smart glasses is relevant. That is good manners. As a system for protecting the public, it fails three times over. It fails on scale: you cannot announce yourself to every passer-by in a station. It fails on incentive: the CHI data shows wearers consistently willing to give less than bystanders want, and growing less concerned with use. And it fails catastrophically on bad actors. The man at the Frog Pond did not intend to tell anybody anything. Any safeguard that depends on the wearer's goodwill protects people only from wearers who were never the danger.
Now the bystander. The research community has proposed plenty of tools to help people defend themselves: gestures that signal “do not record me,” wearable markers that ask cameras to blur you, apps that negotiate consent. The CHI team had experts and experienced users evaluate twelve such privacy-enhancing technologies and found what they called a “fundamental paradox”: empowering bystanders “requires burdening them with responsibilities that should belong to those creating privacy risks.” Participants were blunter. One described gesture-based refusal as “transferring the burden to the victim.” Another asked, of awareness tools in general: “What's the point of awareness? I know, but can I stop it?”
The EDPB reaches the same conclusion from the legal side. Smartphone alerts that detect nearby glasses, it says, “should not be the sole method of providing transparency,” because that would exclude anyone without the app running and would make a stranger's privacy depend on their own technical diligence. Children, it adds, “may be less aware of the presence or implications of smart glasses and therefore more vulnerable.” A toddler at a spray pool cannot install an app.
So the premise holds. If the burden of informing bystanders cannot sit with the wearer or the bystander, it has to sit somewhere else. The EDPB's answer is the manufacturer and the AI service provider. The question is whether that answer actually solves the problem.
Does Moving the Burden to Manufacturers Fix Anything
Here is my position. Placing the burden on manufacturers is the right move and an incomplete one. As the EDPB has framed it, an obligation of transparency, it mostly identifies who is accountable for a harm that will keep happening. It becomes a real solution only to the extent that accountability is turned into hard constraints on what the product can do, and even then a residue of harm is unavoidable for as long as a camera can look like a pair of glasses.
Begin with what the shift does achieve, because it is not nothing. The manufacturer is the only party that can change the product. It decides whether the LED is bright or faint, whether the camera works when the light is broken, whether bystanders' faces are blurred on the device before anything reaches a server, whether footage is stored by default, whether face matching exists at all. The EDPB's recommendations are a menu of exactly those choices: on-device processing, blurring or avoiding processing of bystanders' identities, persistent storage disabled by default for real-time assistance, geofencing in sensitive places, and “privacy frictions” that slow down sharing. Every one of those is a design decision only a manufacturer can make. Assigning the obligation to the company that controls the design is not a technicality. It is the only assignment that matches responsibility to power.
The legitimate interest paragraph may matter even more. If commercial reuse and AI training on bystander data are “particularly difficult to demonstrate” as lawful, then the most valuable part of the footage, its afterlife as training material, becomes the most legally exposed. That attacks the incentive that makes always-on capture attractive in the first place. A company that cannot monetise the street has less reason to see it all.
Now the limits, and they are serious.
First, transparency is not protection. The EDPB's main ask is that bystanders be told. But telling someone they are being recorded does not stop them being recorded, which is precisely the point the CHI participant made. A brighter light turns covert surveillance into overt surveillance, which is an improvement, and then hands the bystander the same impossible choice: confront a stranger, leave, or put up with it. The CHI data suggests most people will avoid the camera, which is the chilling effect in a different form. Perfect transparency would make it clear who is watching. It would not make the street feel unwatched.
Second, manufacturer accountability does not reach the people causing the worst harm. The tampering market shows how quickly a hardware safeguard becomes a $60 problem. And a design rule binds only the companies a regulator can actually reach, which is not the same as every company whose products end up on European faces. The person who films children at a pool is not going to be deterred by a design standard. What the manufacturer framework catches is the mainstream product used by mainstream people, which is valuable but is not where the most serious incidents begin.
Third, the framework is preliminary guidance, not law. The statement interprets the GDPR; enforcement depends on national regulators bringing cases. And where legislators have tried to write hard rules, they have struggled. In California, State Senator Eloise Gómez Reyes's SB 1130 would have required wearable recording devices sold in the state to carry a prominent recording indicator from 2028, banned technology designed to conceal it, and penalised covert recording in private areas of businesses. It passed both chambers comfortably. On 30 September 2026, Governor Gavin Newsom vetoed it, saying its definition of a wearable recording device was too broad and noting that existing law already covers recording in private spaces. TechNet, a trade group whose members include Meta, Google and Amazon, had opposed it. When the EDPB adopted its statement two weeks later, it pointed to the bill in a footnote as an example of legislation under discussion elsewhere.
Fourth, and most fundamentally, there is the residue. Even a perfect manufacturer regime, with unmissable indicators, no tampering, no default storage, no face matching and no training on bystanders, still leaves a world in which some people wear cameras on their faces and everyone else has to think about it. The EDPB's paragraph 8 concedes this directly: people may object “even if the device could be identifiable as smart glasses,” simply because the possibility makes them uncomfortable. That discomfort is not a compliance failure. It is the cost of the capability existing in that form at all. No allocation of legal duties can remove it. It can only decide who answers for it.
So does placing the burden on the manufacturer merely identify who is accountable for an unavoidable harm? Partly, yes. Part of the harm is unavoidable, and we should stop pretending otherwise. But identifying the accountable party is how the avoidable part gets avoided. The honest framing is that the manufacturer burden is necessary but not sufficient, and its value depends entirely on whether regulators treat transparency as the end of the obligation or the beginning.
What would make it sufficient, or close to it? Two things the EDPB gestures at but does not yet demand. The first is substantive limits rather than just disclosure: no identification of strangers, no training on bystander footage without a lawful basis that genuinely exists, cameras that fail closed when tampered with. The second is protection that attaches to places rather than people. Buolamwini's proposal in Nature, “consent-conscious spaces” such as schools, hospitals, courts and swimming pools “in which covert recording is prohibited by law,” is one version. The EDPB's own list of sensitive contexts (medical facilities, schools, changing rooms, swimming pools, beaches, toilets) is another. Hamburg's proposals for pools and public transport are a third. Norway's is a fourth, and the broadest yet. On 5 October 2026 its government said it would ask parliament for a temporary ban on smart glasses in certain public places, starting with spaces where children gather, such as schools, kindergartens, playgrounds and youth clubs, and possibly extending to parks, beaches, museums, shopping centres, concerts and sports events. It is not a blanket ban: private use would still be allowed, and exemptions for vulnerable groups and “socially beneficial uses” are being considered. It is also, for now, only a proposal. But it is a government choosing to protect places rather than waiting for people to protect themselves. The CHI researchers reached the same place from the design side, recommending “automatic protection in sensitive contexts” rather than case-by-case negotiation.
Some of this is already happening without waiting for Brussels. In August 2026, HM Courts and Tribunals Service confirmed that people bringing Meta smart glasses into courts and tribunals in England and Wales “will have them confiscated on entry,” following a High Court case in which a litigant was found to have been coached through his glasses during cross-examination. New York's court system had already taken similar action. Spaces are starting to defend themselves, one door at a time.
The Street Worth Keeping
Go back to the Frog Pond. What protected the children there, to the extent anything did, was not an LED, a privacy notice or a design standard. It was other adults who noticed something wrong and said so. That is Jacobs's “eyes on the street” working as she described. It is also a reminder of how fragile it is. Nothing in the reports suggests his glasses announced anything; witnesses noticed what he was doing. Next time, the device will be smaller, the behaviour more careful, and the footage uploaded before anyone has time to object.
The EDPB statement is important because it refuses the comfortable fiction that the wearer can be trusted to manage this, or that the bystander can be expected to. It points to the only party with the power to change the product. Anu Talus put the principle simply in the board's announcement: “individuals deserve clear warnings, not hidden surprises.” That is right as far as it goes. But a warning is not a refusal, and a street full of clear warnings is still a street where people check themselves before they laugh too loudly or speak too freely.
The real question smart glasses put to us is not who should fill in the paperwork. It is which parts of public life we are willing to let become permanently recordable, and which we intend to keep for the glance that ends in looking away. Manufacturers can be made to answer for the first. The second is a choice for the rest of us, and it is one we are making right now, mostly by not making it.
