I wouldn't be so scared. A few things we've been discussing on the art and writing side of things is regulation and the growing feedback loop that's making AI worse at generating a variety of content.
On the regulation side, we know that all this tech is only possible because it's fed ungodly amounts of stolen human-made content, and that for the most part, the law has yet to catch up with that. With image generation, it's easy to prove that copyrighted material is ingrained into it and will produce plagiarized results even if you don't prompt it by name ("cartoon sponge" almost always gets you SpongeBob, stuff like that) meaning it's a huge liability to use for profit - you could be plagiarizing without knowing it, since it's all automated in a black box trained on incomprehensible scales of theft. This obviously applies to music, too. You won't know that the Automated Plagiarism Machine lifted/replicated entire portions of its generated tune from an existing, copyrighted song unless you recognize it or until you go to market with it and get your pants sued off.
When regulation, lawsuits and legal precedent catches up, and training data for for-profit generators can only be obtained through legitimate channels, new models will essentially be lobotomized versions of their former selves. Music generation will be pretty weak if it can only process stock music and music in the public domain. There'll be homebrew models that still steal in order to generate more contemporary styles, but with advances in AI come advances in AI detection. The effort required to launder generated content into something that can't be recognized as AI will likely (continue to) be more trouble than just being original from the start. Everyone assumes things will only get worse and further out of control, but we're in the wild west phase right now. It's as out of control now as it will ever be, and it can only get more regulated from here.
As for the AI itself, there's already evidence that generators are getting less creative as their own generations start getting fed back into themselves as training data. Language models are the easiest to track for this, but image generators are also ingraining their mistakes and making increasingly uncreative and predictable results. As AI floods the internet with slop, it can't take new scraped data without eating its own slop in the process. It's inbreeding with its own data. It can't tell good content from bad. In order to do that, they need better AI detection, which is a double edged sword for them. Tech like that empowers the majority of people that are broadly anti-AI, so there's really no path forward for AI where it both gets better and less detectable.