At that point the administrator of an instance that doesn’t want to federated with the LLM instance, has to defederate from it, updating their instance and then still getting access requests from the LLM instance when it attempts to do the reply, post, community thing as described before.

That is a cost that, historically, the type of instance that would block an LLM instance just for having AI bots has no problem with expending fourfold.

Since I presume that few users would try querying an LLM instance, and therefore few instances would likely even touch the LLM instance (again, remember, the posts would be unlisted), this is pretty much an imagined problem, even if the hypothetical LLM instance were to exist in reality.