>>>In a legal context where detecting CSAM is a strong requirement, what’s preferable to this approach?
It's not a strong requirement though? Only if the company sees it (unencrypted) do they need to report it, detecting it on-device is wholly different. It would honestly be better if they adjusted their TOS and started scanning the files on upload (on server side). They have the encryption keys already.
Apple is planning to install on all (recent) iPhones software that will let them scan for any image similar (using perceptual hashing) to some set of images. Right now, that's CSAM, and only on upload to iCloud. But what do you want to bet China is salivating at the idea of including images of Tank Man and other "seditious" content? And maybe checking images anytime they're added to the phone, or sent to someone else, not just up to iCloud?
It completely ruins the idea that Apple has your privacy/security in mind.