As such, it should NEVER do anything that isn't in your best interest-- to the greatest extent possible under the law. Your relationship with your personal computer is closer and more trusted than your relationship with your doctor or lawyer-- in fact, you often communicate with these parties via your computer.
We respect the confidentiality you enjoy with your professional agents but that confidentiality cannot functionally exist if your computing devices are not equally duty bound to act in their users best interest!
This snitching 'feature' is a fairly general purpose tracing/tracking mechanism-- We are to assume that the perceptual hashes are exclusively of unlawful images (though I can't actually find a firm, binding assertion of that!)-- but there is nothing assuring that to us except for blind trust.
Even if the list today exclusively has unlawful images there is no guarantee that tomorrow it won't have something different-- no guarantee that some hysterical political expediency won't put images associated with your (non-)religion or ethnicity into it, no guarantee that the facility serving these lists won't be hacked or abused by insiders. Considering that possession of child porn is a strict liability crime, Apple themselves has presumably not validated the content of the list themselves and certainly you won't be allowed to check it. Moreover, even if there were some independent vetting of the list content there is nothing that would prevent targeted parties from being given a different unvetted list without their knowledge.
The pervasive scanning can also be expected to dramatically increases the effectiveness of framing. It's kind of cliche that the guilty person often claims "I was framed"-- but part of the reason that framing is rare is because the false evidence has to intersect a credibly motivated investigation, and they seldom do except where there are other indicators of guilt. With automated scanning it would be much more reliable to cause someone a world of trouble by slipping some indicated material on their device, and so framing would have a much better cost/benefit trade-off.
Any of the above flaws are sufficiently fatal on their own-- but add to it the potential for inadvertent false positives both in the hash matching and in the construction of the lists. Worse, it'll probably be argued that the detailed operation of the system must be kept secret from the very users whos systems it runs on specifically because knowledge of the operation would greatly simplify the malicious construction of intentional false positives which could be used for harassment by causing spurious investigations.
In my view Apple's actions here aren't just inappropriate, they're unambiguously unethical and in a more thoughtful world they'd be a violation of the law.