The Scraping Wars
In 2013 a machine photocopying the index was art. A decade later it was litigation. Death of the author, continued.
Theory, 3 July 2026, 6 min read
In 2013, a machine that photocopied the search index was an art project. You typed a word, the system pulled the top-ranked images into a fixed template, and a book came back in the mail with no author on the spine. The design press wrote about it warmly: It’s Nice That (opens in new tab) treated it as a way to churn out fanzines on whatever your niche interests require.
A decade later, machines drank the entire library: every indexed page, every captioned image, every sentence anyone had left lying around in public. The response was not a review in a design blog. It was litigation. The difference between the two moments is worth taking apart slowly, because it is not the difference between small and large. It is the difference between quoting a culture and digesting one, and somewhere between those two verbs the twentieth century’s entire settlement about authorship quietly came apart.
The first dissolution
Anonymous Press belonged to a tradition that was already old when it arrived: the found object, the readymade, the cut-up, appropriation as argument. When the press set eight search results for “the sublime” into an octavo, it was doing what artists had done with scissors and glue for a century, only faster and with less handwork. The gesture depended on the source material staying visible. You could see that the images were scavenged. The scavenging was the work. The author had stepped back, but you could still see exactly where they had been standing.
Call that the first dissolution: the author dissolving into the index. The word you typed went out into a public commons of images, and what came back wore its origins openly. The press never claimed to have made the pictures. It claimed, mischievously, to have made the choice, and then it disclaimed even that by handing the choice to a ranking algorithm. What was left of authorship was a residue: a word, a template, a timestamp.
It felt radical at the time. It turned out to be a rehearsal.
The second dissolution
The systems that arrived a decade later do not quote the commons; they metabolize it. A model trained on the open web contains no retrievable image of “the sublime”, no page you can point to, no seam where one source ends and another begins. The training corpus dissolves into weights: billions of numbers that remember everything and cite nothing. Ask such a system for a picture and it does not find one. It secretes one, statistically, from the residue of everything it has ever seen.
This is the second dissolution: the author dissolving into the weights. And it is the one that finally made the lawyers stand up. The publishing and stock-photo industries could live with collage, because collage left fingerprints and fingerprints could be licensed. What they could not live with was ingestion without residue. From the Authors Guild’s long war with Google over book scanning (opens in new tab), which the machines won back when they were only indexing, to the wave of suits that includes the New York Times’s case against OpenAI and Microsoft, filed in December 2023 (opens in new tab), the question underneath every filing was the same: is reading at scale still reading? Is a machine that has absorbed your work and can imitate its texture a reader, a thief, or something the statutes never imagined?
The courts have spent the years since splitting hairs that were never designed to be split. In February 2025 a Delaware court found that training a legal search tool on Westlaw’s headnotes was not fair use (Thomson Reuters v. Ross (opens in new tab)); in June 2025 a California court found that training a language model on books was (Bartz v. Anthropic (opens in new tab)). Licenses were invented for a transaction nobody had ever priced, and the word “copy”, the load-bearing word of copyright, the one that assumes an original and a duplicate you can lay side by side, was asked to describe a process that produces neither. The settlement that held from the printing press to the photocopier assumed you could always, in principle, point at the taking. The weights do not permit pointing.
What the press knew early
Rereading the death-of-the-author essay in this light is uncomfortable, in the way that finding an old diary entry that predicted your divorce is uncomfortable. Barthes said the author was a function we could stop performing; the generative system stopped performing it at industrial scale and invoiced the culture for the privilege. The scandal of 2013 (nobody made this book) became the business model of 2023. The joke stopped being a joke when it acquired a market cap.
But the press also knew something its successors preferred not to say out loud: that the commons it photocopied was already an author-erasing machine. The search index never asked the photographers whose images it ranked. Relevance was already a form of ingestion. It was quieter and reversible, but it rested on the same bargain: your work becomes raw material for a system that will not remember your name. The scraping wars did not invent that bargain. They removed its reversibility, and reversibility, it turns out, was the entire moral cushion.
There is a version of this essay that ends in mourning, and it would be dishonest. Something was genuinely lost: the ability to trace a made thing back to a maker, which cultures have used as their unit of gratitude and blame for a very long time. But something was also clarified. The scraping wars forced every writer, photographer and designer to articulate what exactly they thought they owned: the artifact? the style? the labor? the right to be cited? Watch the depositions and you can see a civilization doing philosophy under oath.
Where this press stands
This site runs on a machine that ended up on the far shore of the whole dispute, mostly by accident of design. The engine that composes covers here retrieves nothing, ingests nothing, remembers nothing. A title is hashed into a number; the number seeds a sequence; the sequence draws. There is no training data to subpoena. The library has a title about training data, which is not the same thing. Every form the press produces is derived from the word alone, reproducible forever, owing nothing to any image that ever existed.
That is not a boast about purity. It is a smaller claim than it sounds, and the press makes it precisely: a machine can compose without consuming, and claim nothing it didn’t make. The generative tools now sold to designers cannot say the same. That does not make them evil; it makes them encumbered, and anyone building on them should know what the encumbrance is. (The press keeps an honest accounting of those tools, including what their licensing actually promises, because pretending they don’t exist is its own kind of dishonesty.)
The author has now died twice, once into the index and once into the weights, and both times the obituary was premature in the same way. What survived both dissolutions was not the signature. It was the word: the one decision no system has yet taken from the person at the keyboard. You still have to ask for something. The press has always believed that the asking is the work. Everything since 2013 has only raised the price of believing otherwise: the art, the litigation, the models, and the collapse the models are now feeding themselves.
Published by Anonymous Press without a byline: the work is signed by the system, not the hand. How the press works.
Keep reading
All essaysGenerative Design and the Death of the Author
Barthes wrote the author’s obituary in 1967. The generative system finally cashed the cheque.
When the System Becomes the Designer
On autonomous publishing, and what is left for a designer to do once the machine sets the page.
Language as the Only Design Tool
Anonymous Press hands the visuals to an algorithm. All that remains to the author is the word, which turns out to be everything.