YouTube does this. I have many friends that perform classical piano in their spare time. They record themselves playing a piece that's 200+ years old then put it on YouTube where it gets flagged saying some big record label owns the copyright for it because it's similar to a recording they put out.

This is reinforcing my suspicion that there's gross misunderstanding between creator adjacents and non-creators: Takedowns on free fanfictions never stopped.

It's just many IP holders started incorporating fanfics/parodies as part of their advertisement strategies and started enforcing often unspoken guidelines. There is now a ecosystem or a mental co-dependence between IP holders and creators, and both sides are fine with it. So there be fan contents.

But free fan contents were never "legalized" in the content world, as some seem to assume.

Like many things, I suspect this will end up getting worse before there's a chance for it to get better.

Photoshop does this already, but only if it detects that you're trying to print/create counterfeit money:

https://en.wikipedia.org/wiki/EURion_constellation

it's just ai companies using dirty data and hoping they get away with it - and they do, for the time being, it is a bit trickier to show that 'yep, well that's there', and people don't seem to realize that just using a copyrighted image, at all (downloading, accessing in itself, let alone using for something else), or creating an image that would just "look like" a trademarked character - not "make a 1 to 1 copy" but just "look like" - would be enough for it to possibly be an infringement.

there can be a sufficient fix - taking out potentially infringing images from a dataset, and making an effort to make an actually clean dataset. it's really just a matter of "do you actually have rights to use that content? at all, and in that way". and ai companies continually say 'no...but what if we use it anyway".

and it's kind of a sloppy analogy, because with text to image generators (where you just interact with a model that's offered to you), well - people aren't "uploading copyrighted material into an editor". the copyrighted material is already there in the model, it was used in making of it. and if there was no such copyrighted material that'd fit the prompt, it wouldn't be able to generate something. the infringement lies with the service that uses copyrighted material for a model, and then offers it.

fan fiction and fan works continuously being in a murky area with copyright/trademark is not just a thing of "early days" of internet, it's been there all along and is still very much present. companies could crack down if they wanted, but there is too much of stuff out there, it might be hard to nail down exact people, and it might be plainly not too nice to the fandom. but it is not "impossible", and it is very much not a conversation that ever 'went away' or become "kinda solved" - it isn't.

again, with image editors, text editors, etc. - user is making all the actions with content, and the user would be doing the infringement, in editing and further if they were to choose to publish.

with generative ai - copyright infringement is built into the models. copyrighted works were accessed and used to turn into a model. user is just asking, "is it there". and it is. in some of those demonstrated examples, user is not even asking for a model to infringe on anything but it just does.