Coupling has been the concern for a long time
Privacy International has argued for years that real-time facial recognition should never be coupled with body-worn cameras, and that framing is the useful one. The objection was never to a camera by itself, or to recognition software in the abstract. It was to the join.
When both halves sit inside one procurement, a rule against coupling has something to bite on. Somebody signs for a system, the system has a specification, and the specification either includes recognition or it does not.
The join moved out of the specification
Consumer glasses arrive with the coupling unmade. The manufacturer can state truthfully that the product does not identify anyone, and the statement can be accurate on the day it ships. What the device does is capture and transmit, and what happens after transmission is a separate service with a separate provider, chosen later, by the buyer.
A rule requiring manufacturers not to ship recognition is satisfied in full while the combination exists in practice. Assessed one at a time, the camera clears and the service clears. The thing people are actually worried about is the arrangement, and no assessment ever looks at the arrangement, because no single party assembled it.
Assess combinations, and accept the cost
Our position is that any device whose output can be routed to an identification service should be evaluated as though it carried that capability, because functionally it does. The relevant question is not what the hardware performs on board. It is what the hardware makes easy.
The cost of that standard is real and we would rather concede it than argue past it. Written broadly, it captures almost every camera manufactured today, which is not a workable place to land. The narrowing has to come from something other than capability, most likely from form factor and from whether the person being recorded can tell it is happening. That is a less tidy line than a technical one, and we have not seen anyone draw it well.