You know, as much as I hate it… “you know it when you see I but hard to definet” really is accurate.
There’s plenty of things that aren’t outright illegal that are completely inappropriate around children.
And if there are things that are context specific, it gets a lot harder to make a computer recognize a problem.
Audio cues are easy to scan for and computers are pretty good at recognizing sounds, especially in regards to copyright detection (even if their interpretation of “fair use” clause is still fucked 6 ways)
Video is a lot harder unless the computer is trying to match direct images (it’s a lot easier to recognize a still frame from The Avengers when it’s uploaded full size than it is to recognize a slightly warped, smaller cropped version with someone in front of it commenting on the video)