AI-generated background music and sound effects are good enough for most indie games today, that stopped being the interesting question a while ago. The question that actually costs people money is what the license lets you do with the output once you have it, and that varies wildly between tools.
Read the license, not the marketing page
"Commercial use allowed" on a landing page and "you own the copyright to generated output, unrestricted, forever, including resale" are very different promises, and tools blur the line between them constantly. Before you build a soundtrack around a specific generator, find the actual terms of service section on generated output ownership, not the pricing page, and check three things: can you use it in a commercial product, can you use it in multiple products, and does the license survive if you cancel your subscription.
The training-data question you can't fully resolve yourself
Some AI audio models were trained on copyrighted material without a clear licensing chain, and that's a live legal argument, not a solved one. You can't personally audit a model's training data. What you can do is prefer providers who are explicit and confident about how their models were trained and who indemnify commercial users against claims, because that's the provider taking on the risk instead of you.
Where it's an easy win
Ambient loops, UI sounds, footstep and impact variations, the unglamorous bulk of a game's audio that would otherwise eat a huge chunk of an indie budget, is exactly where AI generation is currently strongest: fast, cheap, and good enough that most players will never notice.
Where you probably still want a human
A game's main theme, the piece of music a player will remember and that becomes the game's identity in trailers and reviews, is worth commissioning from a composer even if everything else in your audio budget is AI-assisted. It's the one asset where "good enough" isn't actually the bar you're aiming for.