I think it is very counterintuitive---even to professional programmers---that you can compute things without letting the computer know them. I literally had to go through the entire paper to understand that this can actually work as long as human in the loop doesn't screw up (see my summary [1], which was revised multiple times at that time).
I'm also closely working together with a team at $WORK who's using a protocol very similar to Apple's but not doing CSAM detection. We are seeing some severe pushback on this technology. I wouldn't be surprised if there are multiple homomorohic encryption based products at Big Tech that have yet to see the light of day.