Somewhere on Threads right now, an artist is quote-posting an AI image with three words: this is theft. Under it, for emphasis: good artists borrow, great artists steal.
Nobody edits the second line out. Nobody catches the problem. The problem is not small, and it is not the only one in the caption.
We love artists. Say that first, because everything after it is going to sound like it forgot, twice. Threads is full of illustrators watching invoices shrink while a model trained on their own portfolio, without a license, without a check, generates a client’s mood board overnight for free. Search a stock site for their real name now, and watch it drown: one company’s generator alone had produced 24 billion images and other assets by mid-2025, and still climbing. That loss has two names. Theft is one. The other is just not being found. Neither is the joke.
But the quote defending against it isn’t evidence either. It’s a confession, and so, it turns out, is the industry being accused.
Nobody in this fight has clean hands. Here are both files, read side by side.
The Line About Stealing Was Stolen
The line usually gets credited to Picasso: good artists copy, great artists steal. Picasso never wrote it down. The closest anyone has traced it, Quote Investigator found, is T.S. Eliot, in 1920, and Eliot’s sentence was stricter than the bumper sticker version. Bad theft defaces what it takes. Good theft, in his words, “makes it into something better, or at least something different.” By 1986 a computing manual had reassigned the line to Stravinsky. By 1988 Steve Jobs was crediting it to Picasso in an Australian newspaper. Eliot got erased, Picasso got the fame, and a sentence about theft got passed hand to hand until nobody remembered whose pocket it came out of.
That’s not a footnote about theft. It’s a demonstration of it, mid-sentence, by the people using the sentence to feel superior.
Three Counts, Read Side by Side
Count one: the scale. For seventeen years, one of the oldest, most trusted galleries in New York sold forged Rothkos and Pollocks to collectors who believed the name on the door over the paint on the canvas. Knoedler closed in 2011. What it sold runs past eighty million dollars, built on a Swiss collector who never existed and a forger in Queens who did.
Universal Music Group sued Udio in 2024 for building a product on its entire recorded catalog without a license. It’s a claim of infringement at a scale no single forger could manage by hand: every song, all at once, no invented Swiss collector required, just a training run.
Count two: the defense that works. In 2013 the Second Circuit ruled that Richard Prince could take another photographer’s book of Rastafarian portraits, paint sunglasses and guitars over the faces, and sell the results at Gagosian as his own. Twenty-five of thirty pieces were ruled “transformative” fair use. The word did the work.
The same word is doing the work again. The New York Times’ case against OpenAI and Microsoft, filed in 2023 and still in discovery three years later, turns on the same question the Second Circuit answered for Prince: does taking someone else’s finished work and running it through a new process count as theft, or as something else entirely? Nobody has answered that for the machine yet. Somebody already answered it for the man with the paintbrush. OpenAI and Microsoft were not contacted for comment for this piece.
Count three: the verdict, or the lack of one. In 1976 a federal court found George Harrison guilty, if that’s the word for something the judge himself said wasn’t deliberate, of “subconsciously” copying The Chiffons’ “He’s So Fine” to write “My Sweet Lord.” That case is closed. It has been closed for fifty years.
Getty Images sued Stability AI on a similar theory, that its models were built on Getty’s copyrighted photographs without permission. In November 2025 the UK’s High Court mostly disagreed, finding no infringement in the model itself. Getty is appealing. A separate US case is still open. Stability AI was not contacted for comment for this piece.
Look: that’s the honest difference between the two rap sheets. The artists’ file is closed cases, decades of them, verdicts already handed down. The industry’s file is still being tried. That is not innocence. It’s an earlier point in the same process, and it’s worth saying plainly instead of pretending both files are equally final, because they are not, yet.
What Theft Actually Turns Into
In 1859, reviewing the Paris Salon, Baudelaire called photography “art’s most mortal enemy,” a refuge for failed painters who could no longer hold a brush honestly. He wasn’t joking. A century and a half later the same warning gets aimed at a different machine, in roughly the same tone, by people who would be furious to be told they were repeating a line.
This is purity framing, both times: define the real ones by who gets excluded, and let the boundary do the work the argument can’t. Why does “we would never” survive as a sentence on either side of this fight? Because saying it out loud is easier than opening the file first. Call it what it is when a field retells its own history and skips the chapters that would change the verdict: historical revision through omission. It runs both directions. The art world skips Knoedler and Cariou. The industry skips Udio’s lawsuit and calls the training set “publicly available data,” which does exactly the work “we have always stood for originality” does on the other side: a sentence that sounds like history and was never checked against it.
Watch what a scandal turns into, once the lawyers get tired. Universal sued Udio for theft in 2024. By October 2025 they’d settled, and now they’re launching a joint AI music platform together, with opt-in compensation for artists built into the deal. That’s not a verdict. It’s a licensing agreement, and it is, not coincidentally, the only version of this fight that ever actually resolves anything. Sony hasn’t signed. Sony is still suing. Both of those are true at once, and neither one is a purity test. They’re just where the receipts stand this year.
Nobody Was Ever Pure. That Was Never the Fix.
Removing the myth of the untouched original doesn’t remove the theft on the other side of the ledger, and removing the myth of the innocent tool doesn’t remove the labor underneath the training run. It removes exactly one argument on each side, the ones that were never going to survive a fact-check: that art was clean before the machine got to it, and that the machine took nothing anyone would miss.
Neither was true. The fix was never going to be purity, for artists or for the companies trying to replace them. It was always going to be the boring thing: consent, a license, a check that clears. Udio has one now. Getty might get one out of a courtroom instead. Either way, that’s the actual mechanism. Not the caption.
A license also doesn’t fix a second problem hiding under the first one: whose work the model learned from to begin with. The training sets skew toward whoever was already easiest to scrape, which tends to mean already famous, already online, already Western. That’s algorithmic discrimination in its plainest form: a bias inherited from the training data, then laundered through the appearance of a neutral process. Artists’ groups and the NAACP have both raised this as a separate harm from theft, not a smaller version of it. That failure needs its own name too. “Licensed” isn’t it.
If you’re the one still deciding, still opening the tool and closing it again, still not sure whether using it makes you a scab or just makes you tired: that’s not confusion. That’s the correct response to an industry that hasn’t settled its own terms yet, and it gets settled by contracts, not by how loudly you refuse a piece of software at your own expense. Nobody here is asking you to trust the machine. Ask what Udio’s artists get to ask now: a license and an opt-in box. If the tool in front of you doesn’t have one yet, that’s the actual thing worth being angry about. Not the tool. The missing box.
We still love the artists, on either side of that decision. We’re not against the tool either. We just want everybody using it, human or otherwise, to say where they got it.
