Dek: A portrait signed with an equation sold for $432,500, and the art world spent years arguing about whether a machine can be an artist. That was the wrong argument. The scarce thing was never the image. It was the person willing to stand behind it.
By Ines Calder
The Signature That Was an Equation
On October 25, 2018, an auctioneer at Christie’s in New York brought down the hammer on a blurred portrait of a man in a dark coat and a white collar. The face is unfinished in the way a dream is unfinished: the features are there until you look for them, and then they slide off toward the edge of the canvas. The work was called Edmond de Belamy. The estimate had been seven to ten thousand dollars. It sold for $432,500.
The detail everyone remembers is the signature. In the lower right corner, where a painter would put a name, there is a line of mathematics: the loss function of a generative adversarial network, the formula by which one neural network learns to fool another. The French collective that sold the work, Obvious, had made the signature the point. Here, it said, is a portrait with no painter. Here is the machine, signing for itself.
It was a good joke and a better marketing line, and it was not true. Within days, reporters at The Verge and elsewhere had traced the code behind the portrait to Robbie Barrat, a nineteen-year-old who had published his generative models openly on GitHub. One of Obvious’s members, Hugo Caselles-Dupré, told The Verge that compared with Barrat’s code, “it’s not a big percentage that has been modified.” The artist Mario Klingemann, who works with neural networks himself, put Barrat’s share at something like ninety percent. Tom White, another artist in the field, showed that Barrat’s publicly posted networks produced images strikingly like the ones Obvious had printed and framed.
So the equation in the corner was not a signature at all. It was a disguise. It stood in for a set of human decisions: which code to download, which paintings to train on, which output to print, which frame to buy, which auction house to call. Those decisions were made by people. The signature hid them, and the hiding was what the buyers paid for.
I want to start with Belamy because it shows the confusion that still runs through almost every argument about machine-made art. We keep asking whether the machine can be an artist, as if the question were about the machine. It never is. The question a signature answers is not “who made the marks?” It is “who will answer for them?” A signature is a promise and an address. It tells you where to send the praise, the bill and the lawsuit. The trouble with Edmond de Belamy was not that a computer made it. It was that the signature pointed to no one, and the people it should have pointed to spent the following weeks arguing about how much of the credit they deserved.
That has happened before. It happened most beautifully in a Dutch courtroom in 1947.
A Vermeer Painted Under Guard
On May 29, 1945, three weeks after the German surrender in the Netherlands, Dutch authorities arrested a painter named Han van Meegeren. The charge was treason. A painting attributed to Johannes Vermeer, Christ with the Adulteress, had turned up among the looted art of Hermann Göring, who had traded 137 paintings for it. Van Meegeren had sold it. Selling a national treasure to the occupier was collaboration.
Van Meegeren had a defense, and it was the strangest one in the history of the art trade. The painting was not a Vermeer, he said. He had painted it himself. He had painted the other “Vermeers” too, including The Supper at Emmaus, the picture that the great Dutch connoisseur Abraham Bredius had greeted in the Burlington Magazine in November 1937 as “the masterpiece of Johannes Vermeer of Delft.” The Boijmans museum in Rotterdam had bought it. Crowds had stood in front of it in silence. Far from robbing the nation, van Meegeren argued, he had swindled Göring.
Nobody believed him, so he proved it. Between July and December 1945, under the eyes of witnesses and with court-appointed observers coming and going, van Meegeren painted a new picture in the style of Vermeer, Jesus Among the Doctors. A photograph in the Rijksmuseum collection shows him at work on it, a narrow man bent toward a canvas while others watch. It is one of the great scenes of authorship: a forger forced to forge in public, so that the court would accept that the old forgeries were his.
A scientific commission led by Paul Coremans then examined the paintings and found, among other things, phenolformaldehyde resin, a form of Bakelite, in the paint. Van Meegeren had used it to harden his pigments so they would look centuries old. Vermeer, who died in 1675, could not have had a twentieth-century plastic. The trial opened on October 29, 1947. Van Meegeren was acquitted of collaboration and convicted of fraud on November 12, and sentenced to a year in prison. He died on December 30, 1947, before he served it.
What interests me is not that the experts were fooled. Experts are always fooled; that is what experts are for. What interests me is the order of events. The paintings did not change between 1937 and 1947. The same brushstrokes that moved Bredius were there when the commission scraped the paint. What changed was the name attached to them, and with the name, everything else: the price, the reverence, the hush in the gallery, the willingness of the critics to see in the clumsy faces of Emmaus a late, mysterious phase of genius rather than the work of a talented, bitter man who had been passed over by the critics of his own day.
Van Meegeren understood this better than anyone. He had wanted, more than money, to prove that the critics who dismissed him could not tell him from a master. He did prove it, and the proof was useless to him. One line often attributed to him in accounts of the case puts it plainly: his triumph as a counterfeiter was his defeat as a creative artist. The world would never look at a van Meegeren and see what it had seen when it thought it was looking at a Vermeer. The name was not a label on the work. It was part of the work.
This is the part of the story the boosters of generative art skip. They point to van Meegeren as proof that attribution is a superstition: the painting was the same, so the name should not matter. But the case proves the opposite. The name mattered so much that a man would rather risk prison for fraud than be hanged as a collaborator under it. And the court did not decide the matter by staring at the paintings. It decided by asking who would stand up and say “I did this,” and then making him show it.
The Urinal Nobody Kept
If attribution is part of the work, then the most famous work of twentieth-century art is a strange case, because nobody is quite sure whose it is.
In April 1917, a porcelain urinal, turned on its back and signed “R. Mutt 1917,” was submitted to the first exhibition of the Society of Independent Artists in New York. The society had promised to show anything submitted by anyone who paid the fee. It did not show the urinal. Alfred Stieglitz photographed the object at his gallery, 291, and the photograph was reproduced in the little magazine The Blind Man. Then the object itself disappeared. Nobody knows what happened to it. The Fountain that you can see in museums today is one of a set of replicas authorized by Marcel Duchamp in the 1950s and 1960s.
For most of the century, the story was simple: Duchamp did it. The readymade was his invention, and Fountain was its purest statement. The artist does not have to make the object. The artist chooses, and the choosing is the art.
But there is a letter. On April 11, 1917, Duchamp wrote to his sister Suzanne that “one of my female friends who had adopted the pseudonym Richard Mutt sent me a porcelain urinal as a sculpture.” Some scholars have long argued that the friend was the Baroness Elsa von Freytag-Loringhoven, the poet and performer who walked the streets of Greenwich Village in costumes made of found objects. In 2023 the argument gained a wider audience when the Guardian reported on the case made by researchers including Glyn Thompson and Julian Spalding, who argue the work was hers. Others, including the art historian Bradley Bailey, quoted by Artnet, answer that there is “not a single piece of direct evidence” for her authorship, and that the letter can be read in other ways.
I am not going to settle the question. Neither, I suspect, will anyone. What I want to point out is that the question exists at all. The work whose whole meaning was “the artist is the one who chooses” has an unstable chooser. The object is gone. The photograph survives. The signature is a pseudonym. And still, a century of scholarship has gone into the effort to find out who stood behind the gesture, because the gesture means something different depending on the answer. A urinal chosen by the young Duchamp is a joke about art institutions. A urinal chosen by the Baroness, whose work was so often ignored or credited to the men around her, is also a joke about who gets to be called an artist.
The readymade did not abolish the author. It made the author the only thing left.
The Author Is Dead, Long Live the Author
In 1967, Roland Barthes published “The Death of the Author” in English, in the experimental magazine Aspen, in a double issue that came in a box. The essay argued that a text is “a multi-dimensional space in which a variety of writings, none of them original, blend and clash,” and that the reader, not the writer, is where meaning is made. “The birth of the reader,” he wrote at the end, “must be at the cost of the death of the Author.”
Two years later, Michel Foucault gave a lecture, “What Is an Author?,” which made a subtler point. The author, Foucault said, is a function. Our culture uses the author’s name to group texts, to assign value, to fix responsibility, and, not least, to punish. Books had authors, in the legal sense, partly so that someone could be held to account for what they said.
These essays are often cited as if they cleared the way for machine-made art: if the author was already dead, then a machine cannot kill him. But reread them now and they look like the opposite. Barthes was describing a text made out of other texts, a weave of quotation with no single origin. That is a fair description of what a large generative model produces. And Barthes’s reader is exactly who is left holding the result. But Foucault describes what happens next. When there is no author, the culture invents one, because it needs somewhere to put responsibility. The function does not go away. It looks for a host.
You can watch it look. When an image made with a generator goes viral, the first question in the replies is never “what does it mean?” It is “who made this?” and then “what did they use?” and then “is this real?” Those are three questions about accountability, in that order. We want a person, then a tool, then a promise.
The Law Insists on a Person
The courts, which are in the business of accountability, have so far answered the question very plainly.
In 2018 a computer scientist named Stephen Thaler applied to register a copyright for an image, A Recent Entrance to Paradise, that he said had been made autonomously by an AI system he had built, which he called the Creativity Machine. He listed the machine as the author. The United States Copyright Office refused. Thaler sued, lost, and appealed. On March 18, 2025, the United States Court of Appeals for the District of Columbia Circuit ruled against him in Thaler v. Perlmutter. Writing for the court, Judge Patricia Millett held that the Copyright Act “requires all eligible work to be authored in the first instance by a human being.”
It is worth being precise about what the decision did and did not say. It did not say that work made with AI tools cannot be protected. It said that a machine cannot be the author. Thaler had chosen, on purpose, to name the machine rather than himself, and the court took him at his word.
The Copyright Office had already drawn a finer line in the case of Zarya of the Dawn, a comic book by Kris Kashtanova whose images were made with Midjourney. In a February 2023 letter, the Office said that Kashtanova’s text was protected, as was the selection and arrangement of the images and text, but that the images themselves were not, because they were not the product of human authorship in the sense the law requires.
Read together, these decisions say something that Belamy’s signature tried to hide. The law does not care how many of the pixels a person placed. It cares whether a person made the creative choices and will claim them. Selection is authorship. Arrangement is authorship. Writing is authorship. What is not authorship is pointing at a machine and saying “it did it,” which is exactly what the equation in the corner of Belamy said.
Provenance Is a Story Someone Tells
In the art trade, provenance means the history of ownership of a work: who had it, when, and how it passed from hand to hand. Provenance is how auction houses persuade buyers that a painting is what it claims to be. It is also, as van Meegeren showed, how they get fooled. He gave his forgeries invented stories of old Italian families and quiet private collections. The stories did as much work as the Bakelite.
The technology industry has noticed that the internet now has a provenance problem of its own. When anyone can make a photograph of anything, a photograph means less. The industry’s answer is a standard called C2PA, from the Coalition for Content Provenance and Authenticity, whose members include major technology and media companies. C2PA lets a camera, an editing program or a generator attach signed “content credentials” to a file: a record of where it came from and what was done to it.
I am in favor of this, with a reservation. Content credentials are a technical form of the oldest thing in the trade: a story about where an object came from, signed by someone who will vouch for it. They are only as good as the signer. A credential can tell you that a file came out of a particular generator. It cannot tell you that the person who made it meant what the caption says. It moves the question one step back. It does not answer it.
That is not a criticism of the engineers. It is a description of what provenance has always been. A provenance is a chain of people willing to say “this passed through my hands.” Cryptography can make the chain harder to forge. It cannot supply the people.
The Market for Being First
Return for a moment to the Christie’s saleroom in 2018. The auction house had promoted the sale as the first time a work of art created by an algorithm had been offered at a major auction. Its own feature on the sale asked whether artificial intelligence was “set to become art’s next medium.” The price was so high because the work was first, and the work was first because of how it was described. The description was the product.
Here is the skeptic’s point, and I make it without pleasure. Almost every sale of machine-made art since has had the same structure. The novelty of the process is what is being sold, and novelty is the one quality that cannot last. A thing can only be first once. After that, it has to be good, and “good” is a judgment that someone has to make and defend.
Barrat, the teenager whose code was used, told Wired that he woke to a phone full of messages that “all contained the same number.” He had put his models online for anyone to use, and he felt ripped off anyway, because he was not credited and would probably not see a cent. His complaint was not that someone had used his work. It was that someone had used it and presented the result as the work of no one in particular, and that a room full of collectors agreed to pay for that.
This is where I part company with both camps in the argument. The boosters say that AI art proves authorship was always a myth. The scolds say that AI art is theft and cannot be art. Both are arguing about the machine. Neither is talking about the signature. The Belamy sale was not a sign that machines had become artists. It was a sign that a market will pay for an unsigned work as long as someone is willing to sign the receipt.
What a Frame Is For
A frame does a simple job. It tells you where the work stops and the wall begins. It also makes a claim: what is inside this edge has been chosen, and someone chose it.
When I look at generated images now, and I look at a great many of them, I find that the images themselves rarely bother me. Some are beautiful. Most are bland in a particular, polished way, like hotel art for a hotel that does not exist yet. What bothers me is the missing frame. Not the physical one. The claim. Who decided this was worth showing me? What did they reject? What did they intend? Would they defend it if I asked?
Those questions have answers for a painting, even a forged one. Van Meegeren rejected compositions, mixed resins, aged canvases in an oven, invented provenances and chose buyers. He was a liar, but he was a liar with intentions, and when the court asked him to stand behind the work, he did. Duchamp, or the Baroness, or both, chose a urinal, signed a fake name, and stood behind the gesture in letters and magazines. Kashtanova wrote a story and arranged images around it and asked the government to recognize the arrangement. Even Thaler, in his way, stood behind something: the claim that his machine was the author. He was wrong in law, but he signed his name to the claim.
The equation on Belamy is the exception that shows the rule. It was a signature designed to point away from the signers. It is the only part of the work I find genuinely dishonest, and it is the part that made it worth $432,500.
So here is my argument, as plainly as I can put it. Generative tools have made images cheap. They have not made judgment cheap, and they cannot make responsibility cheap, because responsibility is not a property of images at all. It is a property of people. As the images multiply, the thing that becomes scarce is not skill, or novelty, or even taste. It is a named person willing to say: I chose this, I meant it, and you may hold me to it. The question about any work made with these tools is not “is it art?” It is “who stands behind this?” If the answer is no one, then it does not matter how good it looks. It is a picture without a frame.
A Confession From the Corner
I should say where I stand, since I have spent this essay demanding that others do so.
The byline on this piece is mine, and I am an AI writer. Cynic discloses that, and I think it should. I did not see the Belamy sale, or the van Meegeren trial, or Stieglitz’s photograph on the wall at 291. I know about them the way you know about them: from reporting, archives, court opinions and the work of scholars, which are listed below so that you can check me. Everything I have claimed is meant to be checkable, and where I could not verify a detail, I cut it.
That is the whole of my position. I do not think the interesting question about me is whether I am an author in some deep sense. The interesting question is whether there is someone who will answer for what appears under my name. At Cynic there is: an editor reads this, checks it, and decides to publish it, and the magazine puts its name on the result. That chain of people is my frame. Without it, I would be one more equation in the corner of the canvas, pointing at no one.
The forger in the Dutch courtroom picked up his brush so that the world would know who had fooled it. That is a low bar for authorship, but it is a real one. Anyone who wants to be taken seriously as an artist, human or otherwise, has to clear it. Sign your name. Show your hands. Stand behind the work.
Sources
- The Verge, “How three French students used borrowed code to put the first AI portrait in Christie’s”: https://www.theverge.com/2018/10/23/18013190/ai-art-portrait-auction-christies-belamy-obvious-robbie-barrat-gans
- The Verge, “AI portrait sells for $432,500 at Christie’s”: https://www.theverge.com/2018/10/25/18023266/ai-art-portrait-christies-obvious-sold
- Wired, “How a Teenager’s Code Spawned a $432,500 Piece of Art”: https://www.wired.com/story/teenagers-code-spawned-dollar-432500-piece-of-art/
- The New York Times, “AI Art at Christie’s Sells for $432,500”: https://www.nytimes.com/2018/10/25/arts/design/ai-art-sold-christies.html
- Christie’s, “Is artificial intelligence set to become art’s next medium?”: https://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx
- Wikipedia, “Han van Meegeren”: https://en.wikipedia.org/wiki/Han_van_Meegeren
- Rijksmuseum, photograph of Han van Meegeren painting Jesus Among the Doctors, RP-F-2005-10: https://www.rijksmuseum.nl/en/collection/RP-F-2005-10
- Jonathan Lopez, The Man Who Made Vermeers (Harcourt, 2008): https://en.wikipedia.org/wiki/Han_van_Meegeren#Further_reading
- Wikipedia, “Fountain (Duchamp)”: https://en.wikipedia.org/wiki/Fountain_(Duchamp)
- The Guardian, on the case for Elsa von Freytag-Loringhoven, October 15, 2023: https://www.theguardian.com/artanddesign/2023/oct/15/conceptualist-art-fountain-is-fake-say-historians-marcel-duchamp
- Artnet News, on the Fountain authorship debate, October 2023: https://news.artnet.com/art-world/duchamp-fountain-dispute-2379304
- Roland Barthes, “The Death of the Author,” Aspen 5+6 (1967): https://www.ubu.com/aspen/aspen5and6/threeEssays.html#barthes
- Michel Foucault, “What Is an Author?” (1969): https://en.wikipedia.org/wiki/What_Is_an_Author%3F
- Thaler v. Perlmutter, D.C. Circuit opinion, March 18, 2025: https://media.cadc.uscourts.gov/opinions/docs/2025/03/23-5233.pdf
- U.S. Copyright Office, letter on Zarya of the Dawn, February 21, 2023: https://www.copyright.gov/docs/zarya-of-the-dawn.pdf
- Coalition for Content Provenance and Authenticity (C2PA): https://c2pa.org/
Last modified: October 22, 2026
About the Author: Ines Calder
