I was on the phone late at night, caking, when I found more than 70 of my songs sitting in two music datasets available to AI developers.
That was not what the call was about.
I was not researching artificial intelligence. I was not thinking about copyright law.
I had come across an article in The Atlantic about millions of songs cataloged in massive music datasets.
There was a search bar.
So I typed in my name.
J. Crum.
Mostly, I was curious.
The first song that came up was “Elevate.”
That was weird enough.
I made “Elevate” when I was deep in my Christian hip-hop career. I believed music had a mission then. I believed God had called me to do this shit.
I toured.
I preached.
I gave the music time, money, relationships and pieces of myself because I thought it was connected to something bigger than me.
Then I kept looking.
LAION-DISCO returned 61 results for my name.
Sleeping-DISCO returned 23.
Eighty-four appearances across two datasets.
I compared the lists later.
Ten songs appeared in both.
After removing the overlap, I had 74 distinct titles.
Seventy-four.
I said something to the person I was on the phone with.
They told me I should lowkey feel honored.
My music was good enough to make it in there.
And for about three seconds, I agreed.
I cannot even lie.
There was a flash of pride.
Like, damn.
They found me?
Then I got irritated.
I did not fully know why yet.
The call kept going for a while, but eventually I got off the phone. Once I was alone, the irritation had room to become something else.
I got angry.
Because why the fuck was I supposed to feel honored?
I have spent my whole career being told I am good at this.
I have won awards. I have opened for artists I grew up listening to. People have cried to my music. People have told me songs helped them stay alive.
I did not need this kind of validation.
For three seconds, I had confused inclusion with recognition.
Nothing had chosen me.
A system had collected me.
What Exactly Happened to My Music?
The easiest version of this article would be:
AI stole 74 of my songs.
That would be dramatic.
It would also say more than I know.
The Atlantic identified four huge music datasets circulating in the AI-development world.
LAION-DISCO contains more than 12 million tracks linked from YouTube.
Sleeping-DISCO contains about 9.7 million tracks, along with lyrics from Genius.
The datasets have been downloaded thousands of times.
Some have documented uses in AI research.
Because companies keep so much of their training data secret, the exact uses of others remain unknown.
So here is what I know.
I found 74 distinct titles across two of those datasets.
I did not know they were there.
Nobody asked me.
Here is what I do not know.
I do not know who downloaded those datasets.
I do not know whether every result was ever retrieved as audio.
I do not know whether a specific commercial AI model trained on my songs.
I do not know whether somebody can already type a prompt somewhere and get back some weird little digital cousin of J. Crum.
The uncertainty is not a footnote to this story.
The uncertainty is part of the story.
And I want to be precise about what that uncertainty means.
A song appearing in a dataset is not the same thing as its audio being retrieved.
Audio being retrieved is not proof that it trained a model.
Those are different steps.
I cannot prove all three happened to my music.
What I can see is the beginning of a pipeline.
I cannot see where my songs went after they entered it.
Three of the four datasets The Atlantic examined are primarily distributed as lists of links to music online. LAION-DISCO itself is an index of YouTube links and metadata, not a giant folder with 12 million audio files sitting inside it.
But developers can use automated tools to retrieve the audio behind those links.
LAION-DISCO alone represents about 91 years of music.
My problem is almost the opposite.
I have to protect my work one song at a time.
I have dealt with unauthorized use before.
Somebody once put my shit on a compilation without asking me.
I found out.
I found the person.
I reached out.
They took it down.
I was irritated, but I understood the world I was in.
There was a thing.
There was a person.
There was an action I could take.
This is different.
Who do I contact about my music appearing 84 times across two datasets?
Who downloaded them?
Where did the data go after that?
If I get a link removed now, does that matter?
If something already trained on it, what would removal even mean?
When I checked again after The Atlantic story ran, public access to Sleeping-DISCO had been taken down.
But taking down a page cannot pull back copies that were already downloaded.
The page can disappear.
The uncertainty does not.
I made the music, but somehow I am still the person with the least information about what may have happened to it.
That is what powerless feels like.
Not simply that something happened.
That I do not even know enough to understand everything that may have happened.
The people operating at the largest scale have the most information.
The person whose name is on the song has the least.
Who Gets to Say Yes?
The strongest argument against me is not that artists do not matter.
It is that machines have to learn from something.
Human musicians learn from music too.
I learned to rap by listening to rappers.
I learned to write by reading writers.
Nobody asked Jay-Z for permission before I studied the way he tells a story.
AI companies have made their own version of that argument.
When major labels sued Suno and Udio for training on copyrighted recordings without permission, the companies argued fair use. They said their systems were built to create new music, not reproduce old songs.
That argument deserves better than a straw man.
There is a real question about what would happen if every developer had to separately negotiate for every single song before building anything.
Who could afford that?
Probably not some kid making something new in a basement.
Probably the same giant companies that already own most of the shit.
A licensing system built badly could turn consent into another gate only powerful people can afford to walk through.
I understand that.
I am not a lawyer. The law here is unsettled, and I do not know how every court will rule.
But legal and consensual are not the same question.
Something can survive a lawsuit and still leave the person whose work helped make it possible with no meaningful choice.
And choice seems to become much more possible once the people demanding it have enough leverage.
The labels sued.
Then the language started changing.
Licensed.
Authorized.
Opt-in.
New revenue.
Universal settled its lawsuit with Udio and announced plans for a platform trained on authorized and licensed music.
Warner settled with both Udio and Suno. In its Suno deal, Warner said artists and songwriters would be able to opt in to certain uses of their names, images, likenesses, voices and compositions.
Sony has not settled with either company.
Universal’s case against Suno is still ongoing.
I do not read corporate press releases as evidence that everybody suddenly found God.
But they prove something simpler.
Permission is not technologically impossible.
Licensing is not conceptually impossible.
Opt-in is not impossible.
Then musicians complicated the story again.
In June 2026, the American Federation of Musicians sued Universal and Warner.
The union alleged that the labels had licensed recordings its members worked on to Suno and Udio without properly compensating the musicians or giving the union enough information to know which recordings and whose work were covered.
Universal disputed that picture and said its licensing work was meant to protect and compensate artists and songwriters.
And there it was.
The question underneath the whole conversation.
Not just:
Did somebody have permission?
But:
Who had enough power to say yes?
The label?
The artist?
The person who owns the master?
The person whose actual voice is on the song?
Who gets negotiation?
Who gets a check?
Who gets a lawyer?
Who gets an email?
And who finds out by typing their own name into a search bar at night?
I am not asking every AI company in the world to personally call me.
I know that is not how licensing has to work because I have already lived through another version of it.
My music has been used by the Seattle Seahawks and the University of Michigan.
Neither one called me personally.
They did not have to.
I had signed a contract with a music licensing company.
That company had permission to represent my work and license it through a system I had agreed to.
There was still a chain of consent.
I said yes to somebody.
Somebody had authority to act on my behalf.
There were terms.
There was a process.
I knew I had entered that system.
That matters.
The Seahawks did not need my personal phone number.
The licensing company had my permission.
That is the difference.
The problem is not that nobody from an AI company called me.
The problem is that I keep looking for the person I authorized to say yes, and I cannot find them.
A major company believes somebody used its property without permission and a whole machine turns on.
Lawyers.
Staff.
Metadata.
Money.
Relationships.
Enforcement.
An independent artist finds 74 distinct titles across two datasets and has to figure out who the fuck to email.
The machine gets scale.
The artist gets paperwork.
I Would Have Said No
The strange part is that I do not have to wonder what I would have said if somebody had asked.
No.
I would have said no.
Not because I think all artificial intelligence is evil.
That would be bullshit.
I use these tools.
I used them on this one.
Not to write it.
To argue with it.
I had AI push on my logic. Ask the questions I was dodging. Check my claims against the record. Point at the places I hide behind a good sentence.
The voice is mine.
The life is mine.
The argument is mine.
But I am not going to pretend I built this in a sealed room.
I am inside the thing I am criticizing.
My hands are not clean enough for purity theater.
That does not make the question of consent disappear.
My use of AI does not retroactively authorize the use of my catalog.
Whether I love the technology, hate it or use it every fucking day, the question is still the same.
Did I get to choose?
I would have said no because I do not trust what happens after yes.
What exactly would I be agreeing to?
One model?
One company?
One use?
What happens if the company gets sold?
What happens when the technology changes?
What happens if permission to study my music becomes the ability to imitate my voice?
If I change my mind in five years, can anybody remove what a model already learned?
Can you prove I am out?
I do not know the answers to those questions.
That is why I would say no.
Maybe somebody else would say yes.
Good.
Let them.
Maybe a young artist needs the money.
Maybe an experimental musician wants to hear what a model does with their whole catalog.
Maybe somebody wants to release every song they have ever made for research because they believe the technology is worth building.
Good.
Their yes should matter.
So should my no.
That is why opting out has never felt like enough to me.
Opt-in asks:
Can we use this?
Opt-out says:
We already did. Let us know if you have a problem.
Those are not different versions of the same thing.
An opt-out system rewards the people who know enough to object.
People with lawyers.
People with time.
People who understand the technology.
People who know what a dataset is.
People who know where to search.
People who discover the use before it is too late for their objection to matter.
Everybody else becomes permission by default.
And I can already hear the question.
How are companies supposed to ask everybody?
There are millions of songs.
Complicated ownership.
Old contracts.
Different laws.
People who cannot be found.
I understand that asking is hard.
I just do not understand why the difficulty of asking becomes the artist’s problem.
The fact that consent requires infrastructure is not an argument against consent.
Build the fucking infrastructure.
We have already built entire businesses around figuring out who owns what because somebody wanted permission to use it.
Those systems are flawed.
Artists get screwed inside them all the time.
But nobody looks at the complexity of music licensing and concludes that ownership should disappear because tracking it is annoying.
A fairer system would not require one perfect contract for every artist and every technology.
But permission would need to be specific.
Permission to use my recording should not automatically mean permission to use my lyrics.
Permission to use my lyrics should not automatically mean permission to imitate my voice.
Permission for one model should not become permission for every future company that model touches.
There would need to be compensation where somebody is building commercial value from the work.
There would need to be some way to trace what was used and where it went.
Yes would need limits.
And no would need to mean something.
I understand why companies do not want that.
Asking is harder than taking.
Tracking is harder than not tracking.
Paying is more expensive than not paying.
Scale makes all of those problems harder.
But scale is the advantage of the technology.
It should not also become the excuse.
You Never Heard Me
The 74 titles I found span a decade of my life.
Some of the music went back to my 2016 album, Black Sheep.
The first result I saw, “Elevate,” came from a version of me who was deep in Christian hip-hop.
But there were a lot of versions of me in there.
CHH Josh.
Married Josh.
Deconstructing Josh.
Divorced Josh.
Hoe-phase Josh.
There were songs from when I believed God had called me to make music.
Songs from when I started wondering whether I believed in God at all.
Songs about relationships.
Grief.
Identity.
Some of the results included songs where I wrote about suicide.
“Cry for Help” was sitting right there in the list.
I had to live through something to make those songs.
That sounds obvious until you see your life flattened into search results.
The dataset does not know what I believed when I made “Elevate.”
It does not know what it eventually cost me to stop believing those things.
It does not know which marriage was ending.
It does not know which version of me was trying to stay alive.
It does not need to.
That was the ugliest thought I had after I got off the phone.
It does not need to care about the content.
It only needs the sound.
I know I have to be careful with a word like plagiarism.
I cannot prove a model copied a J. Crum song.
I cannot prove my music trained a particular commercial system.
But I can tell you what the experience felt like.
It felt like discovering that a decade of my life could become useful after everything that made it mine had been stripped away.
They do not need the grief.
They do not need the marriage.
They do not need the divorce.
They do not need the church.
They do not need Omaha.
They do not need the touring.
They do not need the nights I cried.
They do not need the relationships that got less of me because the music got more.
They do not need to know what any of it cost.
The useful part is the file.
The cadence.
The tone.
The structure.
The drums.
The way I phrase something.
The sound of pain after somebody figured out how to make it rhyme.
I did not sit down one day and invent my voice.
I grew up learning how to name feelings before most boys around me were allowed to admit they had any.
I became a youth group leader at twelve.
I spent seventeen years in ministry.
I preached.
I made Christian hip-hop.
I believed things I do not believe anymore.
Then I had to figure out who I was without the thing I had built most of my identity around.
I toured.
I got married.
I got divorced.
I loved people badly.
People loved me badly.
I tried to survive my own mind.
I left Nebraska.
I moved to New York.
I started over.
All of that is in the music.
Not as metadata.
As consequence.
Looking at those 74 titles, what fucked me up was realizing how much of a life can become useful once you strip the life away.
I lived it.
I made it.
You never heard me.
You just wanted the sound.
Get full access to Amen. Aśe. And All That Shit. at amenaseandall.substack.com/subscribe
