Earlier quoted context omitted.
That's what their story is, but unlikely to be true. All "horrific" pictures will have to be graded by a human, because you have the right to face your accuser, the service providers will not let you near their codebase and AI won't be 100% accurate in the next few years. What they will be able to do with that software is trawl through orders of magnitudes more pictures, filter by probable hits and show those to staf…
Cases reported in the media usually talk about people having thousands of pictures each, I wasn't talking about thousands of different people.
The advantage of an algo detecting illegal content isn't in protecting civil servants, it's scanning everyone without them.