Why would you think that? Every distribution, every view is adding damage, even if the original victim doesn't know (or even would rather not know) about it.

I don't think AI training on a dataset counts as a view in this context. The concern is predators getting off on what they've done, not developing tools to stop them.

Debating what counts as a view is irrelevant. Some child pornography subjects feel violated by any storage or use of their images. Government officials store and use them regardless.

I don't think it's how it works.

Some victims feel this way. Some do not.

I would think there's more possibility of it leaking or being abused in a giant stockpile. Undoubtedly, those training sets would be commercialized in some way, potentially causing some to see that as adding insult to injury.