So this proposal of closed internet and heavily walled gardens doesn't really work and it just leads to more sophisticated and harder to detect means of scraping that data.
Think about it, if you have data that is so valuable that people are willing to put a paywall and throttle their usage at the expense of user experience, its most certainly going to be dumped in the public.
Also this fear of LLM stealing your content is unwarranted. It's no different than somebody running compression algorithms on content they scraped so that its more accessible for the masses.
From my readings and speaking with lawyers it appears the current copyright laws won't stand with LLMs because of the difficulty of reproducing the entire "compression" process which requires extreme amount of money to replicate.
This is OpenAI and many other's legal defence. That there won't be attempts to replicate the underlying mechanism to prove copyright contents were used in the process.
A carefully written prompt to replicate copyright work like with Suno can claim compensation pro delicto but to prove every piece of content will be difficult to claim LLM or token based attention networks are able to "steal" without building it from scratch, training it on millions of hours, especially if its closed and patented.
So we might see people claiming victory over cases like Suno but they've already lost the bigger battle over LLM's legal status.