AI providers have scrapped and will continue to, all internet published information or virtually so. Since "Information" is infinite, AI cannot contain "all information" in a complete sense. But it certainly answers almost everything that matters for any existing search query that has ever been targeted by a webpage that is crawlable.
In any case, as manifest by real world SEO, which is plummeting in traffic for informational queries, the effect is the same. This real world impact is what matters and will not be reversed, regardless of attempts at blocking.
No, it most certainly does not. It was certainly trained on large swathes of human knowledge/interactions.
A model that consists of a perfect representation/compression of all this info is a zip file, not a model file.