That said, to reiterate the OP, there's no evidence that VLU executes when using Finder or QuickLook alone.
"While VLU is taking place, the image being looked up is opened in the floating window of a QuickLook preview."
[0]https://eclecticlight.co/2022/03/23/how-visual-look-up-works...
> Although the original description given was ‘Finder browsing’, for some that might include the display of images as QuickLook Previews, by selecting the image and pressing the Spacebar... [The process triggered from this] is consistent with the briefer task used in Live Text, and quite different from VLU. There is thus no evidence of the generation of neural hashes or any search query by PegasusKit typical of the later stages of VLU...
> Local images that are viewed in QuickLook Preview undergo normal analysis for Live Text, and text recognition where possible, but that doesn’t generate identifiers that could be uploaded to Apple’s servers.
>It’s perhaps worth pointing out here that VLU is nothing like Apple’s proposals for CSAM recognition. The neural hashes used to look up images in VLU are ‘low resolution’. They’re good for distinguishing specific images of paintings, breeds of dog and cat, and so on. But – as you can read in my detailed account of the proposed method – they fall far short of what would have been required to detect CSAM. Another big difference is that initial CSAM detection was intended to be performed locally, using a database of ‘positive’ neural hashes stored locally. Apple recognised that looking them up online wasn’t feasible, but that’s exactly what VLU does, because it works differently, with a much larger database and forming less robust and more inaccurate matches. So VLU ≠ CSAM detection, not by a long way.
I'm still not super sure it couldn't be misused, though. A tool that is good enough to be able to reliably detect things like dog breeds (as the author also points out) seems like it should reliably enough also do things like detect drug paraphernalia, pictures of tankman, banned religious symbols, or whatever thing that people were more worried about being oppressed by a potentially authoritarian government. That's also why it seems, to me, like the more worrying vector for an authoritarian government to force abuse wasn't the higher resolution neural hashes matched against known images but the ML labeling of images for searchability.
It sounds like anytime I open an image in Preview it sends hashes to apple?
"VLU can be disabled by disabling Siri Suggestions in System Settings > Siri & Spotlight, as previously explained[1]."
1. https://eclecticlight.co/2022/04/08/how-to-enable-use-and-fi...
"Images viewed in apps supporting VLU have neural hashes computed, and those are uploaded to Apple’s servers to perform look up and return its results to the user, as previously detailed[1]." (but not for CASM reasons)
1. https://eclecticlight.co/2022/03/25/how-visual-look-up-works...