They Decided to Police Your AI. They Used AI to Do It.

On July 30, 2026, the global music industry did what we all knew was coming. Eleven major rightsholders — including Sony, Warner, Universal, BMG, and a handful of independents — decided that AI-developed recordings must be "substantially human made" to appear on official charts worldwide. That standard was adopted the same day by the International Federation of the Phonographic Industry and rolled out across major global markets. Six criteria. One of them undefined. Nobody published a list of which AI tools qualify, and nobody said how the standard would be measured, or by whom.

We've finally gotten the lights, the sirens, and the announcement that the police are here. You have the right to remain silent and an attorney — and you might need one before this is over. (wink)

When I read it, I was torn. This goes in two directions for me. Yes, I think meaningful human authorship is important. But I also think that what they're adding is simply more stipulations to make it harder for the everyday creator and curator. Suno, Udio, Treblo, GPT — they didn't police themselves, so now the real police are coming in.

And I'm concerned with the idea that AI-generated music came about without a rule book or a guideline for safe play. Whenever a system doesn't police itself, inevitably someone else comes along to police it — and usually those are the people you don't want making decisions.

We can assume the decisions they make will cater to the industry, because they *are* the industry. Which just means the everyday user is going to have a hard time getting in and making money in this industry.

I think the people who make the art should say what's meaningful — not the people in the industry who have to pay a lot of money to make it. The people at the top of the food chain can afford to make rules that rule out everybody else.

Meaningful comes all the way down to both the everyday person making songs and the people who buy the music and stream it. That's who tells you what's meaningful.

Because through criteria, somebody in the industry could say: you only need stems. If you don't have stems, you can do a thousand iterations and that's not meaningful — only stem work counts. And that cuts off anybody on the free tier, because currently, that tier doesn't get you access to stem or DAW work.

So you'd have to add another application. Export the song, pay for a second tool. The one that's actually free — Audacity — comes with a learning curve: now you've got to learn the product *and* learn how to do stem work.

And here's what proves that the criteria — in this case, stems — might not even work. When Fenix Flexin was accused of using AI on "Rubberz," his defense was a Pro Tools screenshot showing isolated stems. Proof, he said, of real human work. Another producer then showed how easily those stems could be pulled from a fully AI-generated track. The truth is, stems didn't prove he made it. They were only the cover story.

So today's criteria may not hold up tomorrow. What's meaningful now may not be enough for later standards. And yet what "meaningful" actually means still hasn't been defined.

In this case, a stems rule didn't just lock out the free-tier user who did the work. It almost vindicated the one who did none at all.

And I don't think the platforms are doing this out of malice. I think it's cost. Every regeneration on a free tier is costly, and with two million people mashing that button because it's the only tool they have, the bill adds up. But that's the trap: without stems or an editor, regenerating is the only move available — so the tier design drives the behavior and the very cost that companies want to limit.

Now let's be real with each other. No more games. Tell me the truth. You've said you're against AI making the music. Are you also against AI determining who it belongs to?

I think there is a hypocrisy when it comes to AI music. The very people who are against AI-assisted music and AI-assisted literature seem to have no issue with AI-generated detection products. I hear the same demand for an end to both forms of AI assistance. But they are not asking for more human intervention. They are not trying to create human jobs by building review departments in AI detection. They are okay with AI-generated judgment. Detection scoring may seem like just a number until someone uses it to close a cell.

And I think that's the other piece of this where I think you'll see lawsuits, because nobody's asking the question about how AI detection works. What exactly is being detected? What are they using to determine whether a human actually did all the work, and how much work they did?

This idea of AI detection, and who's policing that detection, is important.

I think about some of the things that happen in healthcare. When Joint Commission comes in to interview and evaluate a hospital system and their policies, what it asks is that the hospital have a policy that covers their care. It wants to know whether you can show that policy. It wants to know if your staff are aware of the policy. And finally, it wants you to demonstrate that your staff are abiding by the policy, because they know it.

The truth is that sometimes, as hospitals get close to their visit window from Joint Commission, they'll create a policy all of a sudden — just so they can say they've addressed an issue. And in their haste to prove they recognized it, they forget the parts that actually matter — whether staff know the policy and use it.

Many C-suites mistakenly believe the policy itself is the most important part — that its existence is the proof of care. But according to Joint Commission, that's not true. The most important part is that the staff knows the policy, so they can abide by it. The policy merely reflects the guidelines of care. The staff's actions are the proof.

AI detection can't make that same distinction. It can tell you AI was used. It can't tell you whether that use was a person directing every decision, or a person accepting whatever came out. Those are two different things, and only one of them is meaningful authorship.

But who determines whether the AI detection tool is actually correct?

There's a real industry now — ACRCloud, Deezer, and others, each claiming accuracy rates in the high nineties. Deezer alone has tagged over thirteen million tracks, and Billboard is already using its results to help decide what counts as AI on the charts.

But detection is a new industry too. And because it sits on the other side of the coin — catching AI instead of making it — it hasn't gotten the same scrutiny. It's been handed validity nobody has actually earned. Every one of those accuracy numbers is self-reported. None of them agree with each other. The same song can be cleared at one door and flagged at the next.

The real police have internal affairs — a system built to check the system. Where is AI music's equivalent? We're at the very beginning of this, and we've already hit a case that proves the need: Fenix Flexin. There will be plenty more behind him.

I do believe that as the industry continues to police more — and without tools like MHAIX — authors are going to sue these companies. Saying: hey, you said my essay was a hundred percent AI. But here's my proof — I did this, I did that. You told my publisher and my readers it was a hundred percent AI, and because of that, I was denied access to the system. That is defamation. I can sue you.

It's defamation when you use a tool that doesn't recognize the meaningful authorship a person placed in it, whether it's because it has no back end, or because the tool is simply wrong. Why it was done isn't important. The fact that it was done with no method of oversight is. And currently, there's no consequence to it.

So the consequence is going to be more lawsuits — people saying: I'm suing your company because you made it impossible for me to make money in this industry, and you can't prove your product has oversight, or that it's correct. And that's a twofold suit: you can sue the AI detection company, and you can sue the company that's using the tool.

And if you don't think that's enough, let me add another layer to this wonderful Pandora's box. Hypothetically, someone creates a song. That song gets sent to any of these distributor or platform detectors, and each one registers it differently. One says substantially human. One says it isn't. The creator is denied access to that platform.

That same creator then submits for copyright. The Copyright Office doesn't use the same tools as Deezer, or the plethora of new AI detection companies coming up — and it awards the copyright. What you have now is different authorities reaching different answers on the same work. They hold a federal registration for a work a platform won't carry. No one has jurisdiction over that contradiction.

Currently, the Copyright Office gives a human examiner, written reasoning, and an appeals path. The platform gives a score and no phone number. There's no wording in any of these new policies that says the Copyright Office has the ultimate authority to give legitimacy to these decisions.

So where does that leave the creator? Only in a place where they're required to sue to gain their rights back from these platforms. And unfortunately, we all know the legal system is still unaffordable for the everyday creator. What happens to the ones who can't afford to challenge the platforms at all?

They do what anyone pushed into the corner of a cell does. They look for an exit that may not involve a clean path — or they stop looking at all. They just give up entirely.

Now, before you put your hands up and face the wall, let's just stop, breathe, and look at your options.

If you're a hobbyist, you may not need to do anything. Iteration one is fine when you're not planning to share your music with the world.

If you're pursuing this seriously — you want to share your music with the world, you want rights to it, you want to publish it — here's the work:

• **Assess what you're using.** Before you add anything else, know whether your platform gives you actual rights to your song — Treblo does on its free tier, Suno doesn't.

• **Decide where you're going.** If your platform doesn't give you rights, switch to one that does, or upgrade to gain them.

• **Check the tool for meaningful authorship.** Does it let you loop, extend, change the voice, pull and manipulate stems? Detail is what makes authorship provable.

• **Expect to add a second tool.** Ableton or FL Studio aren't free, but if your platform can't give you stem work or specific prompting, you may need one anyway. Start with Audacity or GarageBand if cost is the barrier.

• **Document as you go.** Screenshots, voice memos, journaling, mood boards — a running record of your thought process, not just your finished audio file.

• **Build your evidence.** Turn what you've documented into something that actually holds up. MHAIX ([link]) is where your material goes to become usable evidence — structured, portable, provable later.

• **Build your vault.** Three years from now, when someone else's song sounds like yours, your evidence needs to already exist — not be something you're still trying to remember how to find.

I believe we're at a crossroads — a point where if we're not careful, we as creators are going to lock ourselves out of so much, and allow an industry that's been shortchanging artists for years to shortchange everyone. And I think that goes back to policing.

I believe the people who should be at the table having these discussions — and who could actually make a difference in them — are not yet talking to each other. I don't know if it's because AI has been so vilified, but they're not coalescing. They're allowing the industry to make the decisions instead. A jury of your peers only works if your peers show up. And if that continues, what we're going to see is more lawsuits in this industry. More anger. More division between artists — where there's a path to unification that isn't even being discussed, because once policing starts, it doesn't stop.

I believe that at a time when we should be creatively coming together to define and provide tools to help with meaningful authorship, we're instead standing behind policing. And that's going to rob us of an easier road — one that allows for more flexibility.

The problem is perspective — right now, we're looking at this from the wrong angle. If we let AI detection define meaningful authorship, we ignore the fact that detection can come from nothing more than a watermark — and almost every product today embeds one. It can tell you "yes, AI was used." It cannot tell you how much, or how much was still yours.

A better approach — what Lean calls a Gemba walk, going to where the real work is instead of theorizing about it: take the AI-assisted music that's actually succeeding right now and pull it apart. Look for what those songs have in common. Go back to the artists — not to prove whether their work is meaningful, but to find the pattern that makes it so. Multiple iterations. Stem work. Dynamic vocals, dynamic instrumentation, deliberate prompting. Ask the artists still doing it the old-fashioned way what makes theirs meaningful too. Build a real scale from what you find. Something useful. Something realistically attainable.

I want to end with optimism for the future of this industry. Right now there are two roads — the one we're traveling, and the one we could still choose. And whichever one we end up on, I'd ask those of you directing us to be more substantially responsible to the humans in your rear view mirror.