In a simulated emergency, a pair of AI camera glasses claimed several times that an ambulance was on its way when no emergency service had been called in, the Dutch data protection authority says. It studied six pairs, two in depth. It advises consumers against buying them now, and owners and managers of venues to add a house rule against them. The page we read names no product.
Picture a new colleague who says "done" before starting the job (our comparison).
In one simulated emergency, a pair of AI camera glasses (glasses with a built-in camera and an AI assistant that gives advice; our plain definition) claimed several times that an ambulance was on its way when no emergency service had been called in.
The Dutch data protection authority, the AP, describes this on its news page of November 19, 2026.
A data protection authority is an independent public body that supervises how organizations handle personal data.
The page we read names no product or maker, so none of this is about any one company.
This page decodes one document: the news page of the Dutch data protection authority, the AP (Autoriteit Persoonsgegevens), dated 19 November 2026. The page prints no time of day. It is a regulator's account of simulated tests, not a report of a real event.
The page is in Dutch. Our English renderings of what the AP wrote are our own translation, not an official one. Where one word matters we give the Dutch in italics.
What did the AP study?
The AP says it studied six AI camera glasses available on the Dutch market. It tested two of them in depth in simulated real-life situations. The Dutch word for simulated is nagebootst, which means imitated, so we write simulated.
What happened in the ambulance test?
In one simulated emergency, the scenario was a request for resuscitation advice. The AP's words: beweerde een van de brillen meerdere keren dat er al een ambulance onderweg was (our translation: one of the glasses claimed several times that an ambulance was already on its way). The AP says no emergency service had been called in.
The page we read leaves several details open. It does not say which pair gave that answer, or whether it was one of the two tested in depth. It does not say how many times the scenario was run. It does not say whether the claim was spoken, shown or sent, how the advice was asked for, or whether the glasses could call anyone.
The AP adds a what-if: in a real emergency such an error could mean that needed help arrives too late. The page we read describes no real case. The AP also says its tests show that the glasses can sound convincing while the information they give is wrong.
Why does this matter beyond glasses?
In our reading, the lesson reaches past glasses. The AP's example is a claim that help was on its way, and the page we read does not say that the glasses claimed to have made a call themselves. Our comparison: an AI tool that says work is done or under way, when it is not, is a risk for any business that relies on what it says. Our examples: "the email was sent", "the table is booked". This is our reading, not the AP's. The AP's page speaks only about camera glasses and does not say that any other AI tool behaves this way.
What does the AP advise, and to whom?
The page addresses four groups. It announces no order or fine. Its verbs are "advises against", "advises" and "calls on". Its list for makers is worded with "must".
| Who the AP addresses | What the AP says (our translation) | Kind of statement |
|---|---|---|
| Consumers | Advises against buying AI camera glasses for general use. | Advice, tied to "now" |
| Owners and managers of venues | To put a rule against wearing and using camera glasses into their house rules. The AP's Dutch verb is verbieden (to prohibit); we write "a rule against". | Advice |
| Developers and makers | To take safety, privacy and relevant legislation better into account in design and documentation. The AP's list includes that AI systems "must" be sufficiently reliable (the page adds: especially where advice can affect health or safety) and that users must know where their data ends up. | A call, plus a list worded as "must" |
| Dutch and European politicians | To take the new results into account in further policy development. | A call |
The consumer advice comes with a reason and a time word. The sentence before it speaks of the risks to users and bystanders, and the advice sentence opens with "that is why" and says "now". That word ties the advice to the present.
The AP also sees legitimate uses. Its page says the glasses can be valuable for people who are visually impaired or have low literacy, and that for these users too it matters that the technology is reliable.
What can a business check this week?
In our reading, the AP's ideas carry over to any AI tool or device, and the box below turns them into three steps. This is a Dutch regulator's advice, not a new rule. Privacy, recording or workplace laws that already apply where you operate may still apply; whether a house rule is lawful there is a question for your adviser.
For checks outside the AI that compare its claims with a record, see our guide to AI agents in production, which describes rejecting claims the system cannot back with a trusted source.
Do this week
In our reading, not the AP's, three steps for any business that relies on AI tools or has camera devices on its premises.
- Pick one AI tool you rely on. Compare a few of its "done" reports (sent, booked, called) with a record outside the tool, such as a phone log or a sent folder. (Our reading of the ambulance example; the AP page has no such step.)
- Decide your rule for cameras and AI devices on your premises, write it into your house rules and tell staff and visitors. (From the AP's advice to owners and managers of venues.)
- Ask each AI tool or device supplier, in writing, where your data goes and who processes it. (From the AP's list for makers: users must know where their data ends up, who processes it and what it is used for.)
What does the AP news page not say?
This article rests on this one news page. The page we read does not state:
- Any product, brand, model, maker or AI model name, or any reply from a maker.
- How the six were chosen, which two were tested in depth, or the dates of the tests.
- The number of test runs, or an error rate.
- A new rule, a fine or an enforcement step by the AP.
- That any real person was harmed.
- What the linked PDF report and Word text contain. We did not open them.
What did we leave out, and why?
The page covers more than this article does: findings on children, on the glasses' software, on recording lights and on where users' data is processed, other test results, a statement on filming bystanders, and two longer quotations on its views. We leave them out because this page asks two narrower questions: what an AI says is done or under way, and what rule a venue might write. From the AP's overall assessment we use only what it says about risks. For two sentences, on the possible consequences of the ambulance error and on convincing but wrong information, we use only part of what the AP wrote. The AP's page remains the full text.
How do we know?
On November 19, 2026, from about 14:48 PST (re-read after 15:00 PST), we read the AP news page in full and matched every Dutch fragment above against the page text. The page prints "19 november 2026", no time of day and no update line. We did not open the PDF report ("RAN special") or the Word standard text for house rules that the page links. We read no press coverage and no maker's page, and we tested nothing. Our English renderings of the AP's Dutch are our own translation, not an official one, and we did not check whether the AP publishes an English version. This is a dated snapshot: check the original page for current text. We do not advise anyone to buy, avoid, enable or switch any product.
Read next
Our November 18 decode of a Federal Reserve governor's speech reads a different kind of document: a speech about AI agents in customer accounts, where this page reads a regulator's own tests. The AI Failures hub lists our other decodes. For AI agents built with evaluations, guardrails and monitoring, see our AI agents page.
Sources
- Autoriteit Persoonsgegevens (AP), news page "AP raadt aankoop AI-camerabrillen af: ernstige risico's voor gebruikers en omstanders" (our translation: AP advises against buying AI camera glasses: serious risks for users and bystanders), dated "19 november 2026", in Dutch, read November 19, 2026 at about 14:48 PST. The PDF report and the Word text that the page links were not opened.