In Poland we have a whole catalogue of voided agreement paragraphs, that gets updated regularly. It’s a safety mechanism against companies exploiting the legal system to push people into signing things that are considered unfavourable to them - https://uokik.gov.pl/niedozwolone-klauzule
We’ve had banks for example that had bad terms (or trading platforms), and then government step in and make such terms void.
The reasoning is that a consumer/user should expect fair treatment when agreeing to ToS, without spending money on legal advice before clicking “I agree”.
That is, each legally binging copyright assignment needs an enumeration of fields, like radio, tv, streaming and so on, and when a new field appears the copyright belongs to the original author by default. And statements like “in all fields current and future” are not legally binding.
There is no such thing as “people signed copyright to Meta so Meta can use it in whatever fashion”.
You can read an exercept of Meta’s ToS here - these ToS do not allow them to train models based on the content people publish. https://www.quora.com/When-I-post-a-photo-on-Facebook-or-Ins...
The way I understand it is, the laws are made to cover cases of LLMs being trained on work of authors and artists who normally make a living from the content they create such as articles, books, music, images, videos, etc. whose work is normally copyrighted but LLMs don't give a shit and can just scrape and "learn" your content, style and patterns and then regurgitate it with slight alterations without the original authors getting paid.
It's no secret that some LLMs have been trained on pirated books off libgen.