Lawyers generally say (e.g.KPMG Law/) that code generated by LLMs can not be subject to a copyright by the person who was prompting the LLMs. I believe this bases on an old-fashioned, insufficient understanding of software development.
First let's start with a simple example. This text has been generated by an LLM. The prompt was "Please turn the text below into valid, beautiful HTML with a white background and a black sans-serif font." - I was using the LLM purely for the boring job of HTML formatting. Think about translation to a different language - the creative aspect of a work clearly comes from the human mind.
And coding is not much different. Writing software is not just LLM-style, you sit down, let the mind flow, and voila, code. Instead, the heart of the creative act is:
The rest is just coding / implementation work. The IP of a piece of code happens in a-c, not in when you turn it into code. You don't patent the code, you patent the technical idea, the high-level technical sketch of how your idea solves a problem.
Yes, there are grey areas. For example, in my recent tetris-like retro game, glyph grid/ , I specified precisely how the game shall look like, that it shall use a 3-digit name entry for high scores etc, from which the AI concluded that this should be a retro game and used that info consistently in fonts, colors, screen layout. Has the AI made it a retro game? Or has my spec made it a retro game? Hard to say. But just like when a today's carpenter designs a new table and the machines create it as designed, a software engineer today designs apps, and the machines create it as designed.
Particularly the fact that exactly that what you would prompt for one of the current "thinking" (ie chain-of-thought-reasoning) AIs is what you would patent shows to me that AI-generated apps are copyrightable, and that just the legal system has not yet understood enough about the topic to come to this conclusion.