By that same logic LLMs themselves (by now some AI bro had to vibe code something there)

I'm guessing LLMs are still really really bad at that kind of programming. The packaging of the LLM, sure.

& their trained datapoints

For legal purposes, it seems like the weights would be generated by the human-made training algorithm. I have no idea if that's copyrightable under US law. The standard approach seems to be to keep them a trade secret and pretend there's no espionage, though.