TIDAL announced it would stop paying royalties on AI-generated music. The trade press ran it as a policy on artificial intelligence in music. Read closely, it's a much narrower thing than that, and the gap between what the headline says and what the technology can actually do is where a lot of working musicians are quietly getting nervous.

We had a conversation recently that made the gap concrete.

‍

The artist who mastered with a machine

‍

We spoke to an artist. He plays his own guitars. Sings his own vocals. Picks up a ukulele. He's written hundreds of songs across a long career, all of it his. At the final step, mastering, he uses an AI tool. It saves him a studio bill and it speeds up the part of the process he doesn't consider the creative part.

His question to us was simple and slightly worried. Is he now an AI-generated artist? Are his royalties at risk?

It's worth taking that question seriously, because he is not an edge case. There's a whole cohort of established writers, people in their sixties and seventies with hundreds of original compositions, who are bringing AI assists into one narrow part of the process. Not the songwriting. Not the performing. The final polish, the bit that used to mean booking a room and paying an engineer.

So where does he actually stand?

‍

What the detection can and can't see

‍

Here's the part that resolves most of the worry, and it's worth being precise about it.

The bans that platforms are rolling out target wholly AI-generated tracks. And the reason they draw the line at 100% is not because someone decided that was the philosophically correct place to draw it. It's because that's the only line the technology can currently hold.

Detection works on output signatures. The tools look at the finished audio and try to work out whether the whole thing came out of a generative model. They are not looking at your production chain. They can't see which plugins you used. They can't tell that your master went through a machine learning tool while your performance came off a real guitar in a real room.

Which means the artist mastering his own playing with an AI tool is, under current policy, fine. The detection isn't built to catch him, and it couldn't catch him if it tried. The song is a human performance. That's what the finished file shows.

‍

The Ableton problem, and why it's moot for now

‍

The version of this fear that gets people really spun up goes like this. If the hard line is "anything with a machine learning model in it," then half the plugins in a modern session are at risk. Stem separation uses ML. Plenty of mastering tools use ML. Some of the standard stuff shipping inside Ableton uses ML. On that reading, nearly every record made in the last few years has an AI component somewhere in it.

That reading is right about the tools and wrong about the policy. If "contains a machine learning model" were the standard, the standard would catch almost everything, which is another way of saying it would be unenforceable. No platform can run that line, because no platform can detect it.

So the Ableton concern is real as a principle and moot as a practical matter. Under what the technology can actually do today, the line sits at wholly generated tracks, because that's the only line detection can defend. The plugin-by-plugin panic is answering a policy that doesn't exist yet.

‍

Where the grey zone actually lives

‍

None of this means the question is settled. It means it's unsettled in a different place than the headlines suggest.

The genuine grey zone isn't the artist using an AI master on a human performance. It's the stuff in the middle. A track where the vocal is real but the instrumental was generated. A song written by a person and then substantially produced by a model. Partial generation, where you couldn't honestly say the machine wrote it and couldn't honestly say it didn't. That's the territory nobody has a clean answer for, and it's the territory the detection can't cleanly sort either.

The industry hasn't addressed that middle. It's drawn the one line it can hold and left everything short of it undefined. So if you're a writer wondering whether your royalties are exposed, the honest answer is: probably not today, and it depends entirely on how much of the actual music came out of a model, not on whether a machine touched the file at any point.

‍

Why we care about getting this right

‍

We work with catalogue owners, a lot of them established writers with decades of compositions behind them. When one of them asks us what our view on AI is, "it depends" isn't good enough on its own. They need to know where they actually stand.

Our view has to stay tuned to whatever the policy is actually doing, not to whatever the headline implies. Right now the answer for the artist mastering his own playing is that he's on the safe side of the only line that exists. That might change. If detection gets better, or if a platform decides to draw the line somewhere new, the ground moves and we'll be watching it move.

But panicking about a policy nobody can currently enforce helps no one. The man with the ukulele and the hundred songs is a songwriter who used a tool. That's what he was. That's still what he is.

‍