Apple has declared their criteria for evaluating the privacy of their CSAM scanning approach in their USENIX talk, and none of it relies on scanning all files. If you want to accuse them of lying, you will need to bring a lot more evidence since (i) presumably you believe they weren’t lying in the past and (ii) it would be an enormous liability if they were now.

There are many reasonable criteria under which the proposed model is superior for privacy. Perhaps the only reasonable criteria under which it is not is some kind of scope creep in what files are scanned. (Scope creep in what content is scanned for is a risk - an even greater one as there is no transparency over the hash list - of incumbent solutions as well.)