Fediverse

There are some lawsuits in motion about this and the early signs are that it is indeed legal. For example, in Kadrey et al v. Meta the judge issued a summary judgment that training an AI on books was "highly transformative" and fell under fair use, and similarly in Bartz, Graeber and Johnson v. Anthropic the judge ruled that training an AI on books was fair use. I always expected this would be the case since an AI model does not literally contain the training material it was trained on, it learns patterns from the training material but that's not the same as the literal expression of the training material. Since the training material isn't being copied there's nothing for copyright to restrict here.