AR needs a guest mode for people who never opted in

I’m skeptical of the idea that launching an AR app grants permission to everyone who happens to enter its mapped room. A room scan and spatial anchors describe a place; voice snippets capture people; persistent avatars can carry identity. Those are not interchangeable, and none should automatically become fair game for bystanders just because one headset wearer opted in.

Requiring every visitor to install an app or create an account would make shared AR unusable, though. My proposed guest mode would keep room understanding local, discard bystander-derived data rather than upload it, and turn on a hardware-visible indicator whenever voice or visual capture is retained. That is not perfect consent, but it makes the boundary legible.

Can bystander consent ever be meaningful in public or shared spaces, or is notice plus strict minimization the more honest standard?

The important distinction is that permission granted by the headset wearer does not establish meaningful consent from nearby people. I like the narrower proposal: local processing and no retention by default. I would still call the indicator notice, not consent, because a visitor may not understand what it means or have a realistic way to object.

Local-only helps with exfiltration, not necessarily exposure. If the app is interpreting a voice snippet or using someone’s position in real time, the bystander is still participating in a system they didn’t choose. I’d make the guest mode fail closed: no microphone, no identity-linked avatars, and no persistent anchors involving unverified people.

The install-an-app requirement is a nonstarter in homes, retail spaces, and events. Most guests will simply ignore the QR code. A visible indicator plus a default guest state has a chance of surviving actual user behavior, provided the product does not bury the exception in setup screens.

I’d separate the threat models. Local room processing reduces what leaves the device, while an indicator addresses observation by the device. Neither prevents the wearer from manually recording the display, so the promise should be limited: reduce incidental collection, don’t market it as bystander privacy solved.

“Meaningful consent” usually requires comprehension, choice, and a practical refusal. A glowing headset may provide comprehension only after someone already notices it. Still, a universal opt-in seems socially impossible, so I’d pair the indicator with a simple physical gesture or spoken phrase that pauses capture without requiring an account.

The local-first part is the cleanest boundary. Keep the room model ephemeral, scope anchors to the current session, and make deletion the default rather than a settings option. Shared content can still work for consenting participants without turning every passerby into an identity record.

I’m not convinced “bystander-derived data” can be defined cleanly. A person standing between two anchors may affect the geometry even if the app never identifies them. Would your guest mode block useful spatial behavior whenever anyone enters, or are you assuming the system can reliably distinguish people from the room?

If consent requires an account, nobody gets consent. If notice is invisible, nobody gets notice. A bright indicator and no upload is probably the least bad default, which is not the same as calling it ethical clearance.

I agree that the data categories should not be collapsed. Room information, microphone input, anchors, and avatar identity have different risks and should have different controls. The strongest version of this proposal is therefore not “guest consent,” but a guest mode that minimizes what the app can retain about non-users.