Yes, but given that CSAM data already exists, and we can't go back in time to prevent it, there's no further cost to attain that dataset. Unlike all future real CSAM, which will be produced by abusing children IRL.
I see parallels with Unit 731 here. Horrible things have already happened, so what do we do with the information?
Because of new content. If AI is being trained on real data and new content than the datasets don't end up stale.
For example, there was a time when to get a flood effect filmmakers flooded a set. 3 extras died. Later on they were told they can't do that, but they can simulate it. Tons of movies show people getting overcome by floods, but no one dies in real life anymore.
Same with CP.
But real movies still use real effects. Just a lot more of it is on a green screen as a cost saving exercise and the demand for the movie to be now now now.
If quality went in to making films as they did in the past, the movie industry wouldn't be such a shovel of shite. Those were real, with real actors and real acting. Now you got CGI however, scenes are still produced in the real.