Framing as “on device scanning” is contrary to facts. It’s the client SDK operating the same algo on an upload event image at the client end of the upload API as it would run at the server end, just not having to decrypt it to do it.

But to your theme — there’s no clear bright line in client SDKs.

Devs do not think of these or handle these as if they belong to the user.

If you study the cesspool of user tracking and other privacy violations on iOS devices, it’s mostly developers using the same client-side SDKs across their apps, and users none the wiser.

Sometimes devs also have no idea. At least, they profess surprise and drop these SDKs when the user behavior harvesting is pointed out to them.

For whatever reason, everyone’s OK with this.

Perhaps the line is too blurred between client and server, thanks to emergence of concepts like node.js (“run same code on client or server”) that make this sound like a feature.

Or maybe it really shouldn’t matter where the code runs provided it respects boundaries, or better, those boundaries of intentionality and access are technically guaranteed inviolate.