The issue to me is that the LLMs used to generate said parts are trained on code from all of the internet without regard for licenses or legal restrictions—in short: it is stolen code.

I am of the opinion that this goes directly against copyleft ideals, like those of the GPL etc., by allowing corporations to take software under "restrictive" licenses* and exfiltrate them as proprietary, copyright software.

* i.e. ones that would require them to actually contribute to FLOSS instead of leaching off of it