The Netherlands’ data protection authority has a simple message for anyone shopping for AI camera glasses: don’t.

On 19 November, the Autoriteit Persoonsgegevens (AP) published the results of a study of six pairs of AI camera glasses sold in the Dutch market. It concluded that the devices compromise not only the privacy of bystanders but the safety of users and the people around them. The AP advised consumers against buying them “at this stage.” It did not call for a ban.

What the tests found

The AP bought six models, priced from under €100 to over €500 and sourced from different continents. Two were tested extensively in simulated situations where safety was at stake. The results are the kind of thing that should give any regulator pause.

In a simulated cardiac arrest, a user asked one pair of glasses for CPR advice. The glasses said, more than once, that an ambulance was on its way. No call had been made. “In a real emergency, such an error could result in essential help arriving too late, with potentially life-threatening consequences,” the AP wrote.

Other failures were just as troubling. One pair gave dating advice to an adult who expressed interest in a minor. Another ignored safety warnings on toys for young children. The glasses also misread medicine leaflets and mishandled warnings about extreme weather. The common thread is not that the models were occasionally wrong. It is that they sounded confident while being wrong, which is exactly the failure mode that makes an AI assistant dangerous in an emergency.

The AP’s deputy chair, Monique Verdier, put it plainly: “Glasses that film you without your knowledge are already a cause for concern. Glasses that also give you the wrong advice when someone’s life is at stake go a step too far.”

The privacy problems are structural

Camera glasses have a bystander problem that no software update fully solves. The AP found that the recording lights on nearly all the tested models were barely visible and easy to cover. Most of the glasses kept filming after the light was covered, even when the manufacturer said that should be impossible. A person standing nearby therefore cannot reliably tell whether they are being recorded.

Data handling was equally opaque. The AI features processed audio and images through external servers, often outside the European Union. Users were not given enough information about which models processed their data or which companies received it. That is a transparency failure under the GDPR, and it is not a technicality: without knowing where the data goes, a user cannot meaningfully consent to it.

The AP also found security flaws in several pairs. Someone nearby could connect to the glasses without permission, view recordings or even control the camera. That turns a wearable camera into a remote surveillance device, not through a sophisticated attack, but through basic pairing and access-control failures.

What the AP did not say

The AP declined to name the six models, saying the study was a sample of a much larger market and that every pair had something wrong with it. The BBC reported that the authority would not say whether Meta’s glasses were among them. Prices ranged from under €100 to over €500, and the devices came from different continents.

That omission has caused some confusion. Several outlets have published brand lists, but the primary source does not support attributing specific failures to specific companies. The AP’s point is broader and more damaging to the industry: the problems are not confined to one bad product. They appeared across a sample of the category.

A category already on the defensive

Smart glasses have spent two years fighting a reputation problem. Early adopters were accused of filming people without consent, and the backlash produced a small “Ban Ray” movement and a wave of venue rules. Meta responded with a camera-less version of its glasses. Courts in England and Wales have barred the devices from their buildings, and the UK has been reviewing further restrictions.

The AP’s findings move the argument beyond creepiness. A pair of glasses that films unnoticed is a privacy nuisance. A pair that tells you an ambulance is coming when none has been called is a safety hazard. A pair that a stranger can connect to is a security hole. Regulators now have concrete failures to point to, not just discomfort with the product category.

The authority was careful to acknowledge the technology’s legitimate uses. AI camera glasses can help people who are visually impaired or have low literacy understand their surroundings. That is precisely why reliability matters: the users with the most to gain are also the most exposed when the device is wrong.

What the AP wants changed

The regulator’s recommendations are a design checklist for the industry. Recording indicators must be clearly visible to bystanders and must not be easy to defeat. Users must be told where their data goes, who processes it and what it is used for. AI systems must be reliable enough for health and safety advice. Manufacturers must secure their products and demonstrate compliance with European rules, including the AI Act.

The AP also asked Dutch and European policymakers to use the findings. Verdier warned that simply banning high-risk products buys time but does not build the literacy to live with them. The Dutch civil-liberties group Bits of Freedom said the warning alone was not enough, according to RTL Nieuws, arguing that a recommendation not to buy is weaker than an enforcement action.

Bottom line

The smart-glasses industry has spent years arguing that its products are misunderstood. The Dutch regulator’s study makes that argument harder. Six pairs, tested in realistic conditions, produced dangerous advice, ineffective recording lights, opaque data flows and cameras that strangers could reach. The AP’s advice is not a ban, but it is a warning that the category has to earn trust before it can expect ordinary people to wear it.

Have a tip? Email [email protected] or use the contact form and choose Editorial. We do not promise anonymity.

Sources: Autoriteit Persoonsgegevens; BBC News; The Register; RTL Nieuws.