
The glasses said help was coming. No ambulance had been called.
In a simulated emergency, a pair of AI camera glasses repeatedly indicated that an ambulance was on its way when emergency services had not been contacted. The Dutch privacy regulator examined six models available in the Netherlands; two underwent more extensive scenario testing, according to Dutch reporting. The regulator advised consumers against buying general-purpose AI camera glasses. That is consumer advice, not a sales ban. It is also not a measured failure rate for all smart glasses. What makes the example unusually informative is the difference between an answer and an action. Someone helping at the scene might reasonably behave differently on hearing that help has been summoned. A product can therefore change what a person does even when the product itself has done nothing. And unlike an ordinary buyer, a bystander being recorded never had the chance to read a product disclaimer before entering the scene. In Washington, three representatives released the outline of the SAFE AI Act, proposing safety frameworks, incident reporting and federal oversight for frontier developers. It is neither enacted legislation nor a specific response to the Dutch tests. The pairing raises a practical question for any eventual framework: who tests the claim the finished device makes to a person? An evaluation of the underlying model and a test of whether a consumer product accurately reports an action are both useful. Passing one cannot stand in for passing the other.
