Both cases are real, and neither is complete

The case for keeping the hardware available rests on something we have written about directly. A camera at eye level, paired with on-device interpretation, turns visual information into spoken information without occupying a hand. It reads a sign while someone holds a cane. It describes a room while they carry a bag. The same relocation that worries a recording-lights-and-bystander-notice">bystander is, for a blind or low-vision wearer, the thing that makes the tool usable at all instead of merely available.

The case against it is everything else in this series. A phone raised to record announces itself through effort. Glasses remove the effort and the announcement together. The Irish and Italian regulators flagged this as early as 2021, when they questioned whether a small light had ever been shown to give adequate notice in real conditions, and nothing shipped since has settled that question in the light's favour. Add a face-search service to the same camera, as two students proved was trivial in 2024, and the device stops being a recorder and becomes an identifier, a capability no single product specification fully owns because the recognition sits outside the glasses entirely.

Both arguments are correct. That is the actual problem. A framework that only holds one of them produces a bad answer regardless of which one it picks.

Where the simple answers break

Calling the device neutral ignores that a manufacturer sets the default recording length, whether capture continues once the indicator is blocked, and how fast raw footage reaches a platform or a reviewer. Those are not neutral defaults. They are the conditions abuse either finds easy or does not, and the manufacturer chose them.

Calling for an outright ban breaks somewhere less obvious and more uncomfortable. Any exception written for accessibility has to define who qualifies, which means asking a disabled person to prove something about themselves before buying a consumer product. A prescription, a registration, a visible marker. Every option in that list either excludes someone whose need is real but undocumented, or turns an ordinary wearer with a cane or a guide dog into a subject of public verification. A distinctively coloured frame meant to signal an approved exception does not tell a bystander anything about whether the camera is active. It tells them the wearer's disability status, which was never a bystander's information to have.

What actually needs regulating is the mode, not the object

A camera on a face performs at least four different operations, and they carry entirely different risk. Local interpretation converts a momentary view into speech and discards the image. Remote assistance sends a live feed to one chosen person for a bounded call. Durable capture saves a photo or video that can be stored, edited and redistributed. Biometric identification turns a face into a template and matches it against a reference set, which is the only one of the four that removes a stranger's anonymity outright.

Treating all four as one category gives the lowest-risk mode too much restriction and the highest-risk one too little. A rule should instead follow the operation: local interpretation stays on-device with no retention by default, remote assistance has to make its destination legible, durable capture needs an indicator that genuinely cannot be defeated while the sensor runs, and biometric identification of a stranger gets a separate, far stricter line, the one the ACLU and 75 co-signers already drew around NameTag.

Notice belongs in this framework too, and it is the one part we can point to as already built. A signal that rotates its identifier, excludes anything traceable back to the wearer, and only ever answers whether a device is present, not who owns it, is what Bluetooth awareness without surveillance means in practice, and it is the layer this whole framework has been missing a working example of.

Responsibility can then sit where control actually sits. The wearer controls the act. The manufacturer controls the sensor, the indicator, and the defaults. The service provider controls what happens to the data after capture. The venue controls its own rooms. No single actor was ever going to close this gap alone, and asking one of them to has been the mistake underneath most of the proposals in this series, including some of ours.