The world's first copyright law was a printers' monopoly bill wearing an author's coat. Both of those descriptions are true, and that is the most useful thing anyone writing rules for AI content could know.
In the first years of the eighteenth century, the most powerful lobby in England was a guild of printers, and they had a problem. The Stationers' Company of London had held, since 1557, a royal monopoly over the printing of books, a perpetual and inheritable right to copy that made its members rich and everyone else a pirate. Then in 1694 Parliament let the Licensing Act that underpinned that monopoly lapse, and the Stationers spent the next fifteen years trying to get it back. They petitioned, and failed. They petitioned again, and failed. By one accounting they tried and lost roughly a dozen times between 1695 and 1709, always asking for the same thing: restore our monopoly over the trade.
Then they changed the sentence. Instead of asking Parliament to protect printers, they asked it to protect authors. And in 1710 that bill passed. It was called the Statute of Anne, it is universally described as the world's first copyright law, and it is the founding document of the idea that a creator owns their work. It was also, in its origin, a printers' monopoly bill wearing an author's coat, and the fact that those two descriptions are both true is the most useful thing anyone building rules for AI-generated content could know.
We are, right now, having the 1710 fight again, about a new copying technology, and we are having it badly, because we have noticed only one of the three things the Statute of Anne actually did. The full story of that law is a startlingly complete guide to the questions AI raises about who owns what a machine makes, and it warns us about the one we are most likely to get wrong.
The Statute's first move was to vest ownership in the human author rather than in whoever operated the press. Before 1710, the "copy" belonged to the Stationer who printed it; the Statute was, in the historians' phrase, the first time copyright was vested primarily in the author rather than the publisher. Ownership followed the human creative act, not the machinery of reproduction.
That is precisely the move the AI debate has already made, and it is nearly the only part of this it has gotten settled. The United States Copyright Office holds that copyright requires human authorship; output generated purely by a machine, with no human creative contribution, is not registrable. The test case is almost comically on the nose. Dr. Stephen Thaler built an AI system he called the Creativity Machine, had it autonomously generate an image titled "A Recent Entrance to Paradise," and tried to register the copyright with the machine listed as the author. The Office refused. Thaler sued, and lost, and appealed, and on March 18, 2025, the D.C. Circuit affirmed the refusal, holding that human authorship is required as a matter of statutory law. This March, on March 2, 2026, the Supreme Court declined to hear his further appeal, leaving that ruling standing. So the question everyone treats as the great unresolved frontier of AI, can a machine own what it creates, is not only answered, it is answered all the way up to the top court, using a principle a printers' guild smuggled into English law in 1710. Only a human author holds copyright. The panic is three centuries late.
Where it actually gets hard is not the machine alone but the human and the machine together, and here the historical parallel gets quieter and the modern law gets interesting. In 2023 an artist named Kris Kashtanova made a comic book called "Zarya of the Dawn," writing the story herself and generating the images with Midjourney by feeding it prompts and selecting from what came back. The Copyright Office, in a February 21, 2023 decision, did something careful: it registered the human-authored text, and it registered Kashtanova's selection, coordination, and arrangement of the images, the creative choices of which pictures to use and how to sequence them, while explicitly excluding the raw Midjourney images themselves from protection, on the grounds that they lacked sufficient human authorship. One work, split down the middle. The words and the arrangement are yours; the pixels the model produced are nobody's. That is the genuinely new problem, the assisted middle where a single artifact is part-protected and part-public, and the blunt author-or-not-author instrument handles it awkwardly. It is the only part of Move One still under construction, and it is where the real work is.
Now the part almost no one writing about AI and copyright mentions, and the reason the origin story matters. The author-centric framing of the Statute of Anne was, substantially, a lobbying tactic. The Stationers did not have a change of heart about the dignity of writers. They had a losing hand asking for a printers' monopoly, and a winning hand asking to protect authors, because "protect the poor author from piracy" is a sentence Parliament could vote for and "restore our lucrative monopoly" was not. The rights the Statute created were real, and authors did gain something they had never formally held. But the machinery was built to route most of the practical benefit back through the publishers who bought up authors' rights, and the reframe was the thing that got a stalled monopoly bill across the line. A rule sold as protecting creators can be, underneath, a rule about who controls the means of distribution.
Hold that lesson up to the current fight and it should change how you read every press release. The debate over AI-generated content is conducted almost entirely in the language of protecting human artists, and much of that concern is genuine and earned. But the same language is available to whoever wants to control the platforms, and the entities with the most to gain from a particular shape of AI-content rule are frequently not the artists. When a large AI company argues for a regime of licensing and permission around AI output, ask the 1710 question: is this protecting the creator, or is it building the toll road and calling it a shield? The AI companies are the plausible new Stationers, the incumbents most able to turn a rule written in the creator's name into a moat they own. This is not a claim that every artist-protection proposal is a con; it is the Statute's own hard-won lesson that the author's coat fits a monopolist perfectly well, and you have to check who is wearing it.
The third move is the one the Statute of Anne is actually remembered for by people who study it, and the one the AI debate has almost entirely forgotten. The Statute's durable contribution was not that it assigned ownership. Lots of legal regimes assign ownership. Its radical move was to make ownership expire. A copyright under the Statute ran for fourteen years, renewable once for another fourteen if the author was still alive, with a one-time twenty-one-year term for works already in print. After that, the work belonged to everyone. It entered what we now call the public domain, a concept the Statute effectively invented by putting a clock on a right that the printers had always insisted was perpetual.
The printers fought this ferociously, in a thirty-year campaign the historians call the Battle of the Booksellers, arguing that the statutory term merely supplemented a perpetual common-law copyright they had always possessed. In 1774, in Donaldson v. Beckett, the House of Lords settled it: there was no perpetual right, copyright was the deliberate and limited creation of the Statute, and when the term ran out the work fell into the public domain, full stop. The single most consequential idea in the history of copyright is not "creators own their work." It is "creators own their work for a while, and then the public does."
The AI debate is stuck entirely on the ownership question and has barely begun the limits question. We argue endlessly about whether a prompt-driven image can be owned and almost never about the shape of the public domain in a world of machine-assisted creation: how long the human-authored parts should be protected, how the un-ownable machine-generated parts interact with a body of shared culture, what it means for learning and creation when an infinite supply of AI content sits in an uncertain ownership limbo. Move Three is where the Statute did its real public work, and it is the move we have not made.
Everything above is downstream of a purpose the Statute of Anne stated in its own first line, which almost no one quotes. Its full title begins: "An Act for the Encouragement of Learning, by vesting the Copies of Printed Books in the Authors." Read that construction carefully. The goal is the encouragement of learning. The vesting of copies in authors is the method, introduced by the word "by." Ownership was never the point of copyright. It was the instrument, and the point was a flourishing public culture, more books written, more knowledge spread, a richer commons. Author reward was the carrot; learning was the harvest.
The modern conversation about AI content has inverted this completely. It treats ownership as the end in itself, a thing to be maximized and fought over, and it has lost the sentence that made ownership legitimate in the first place, which is that ownership is worth granting only insofar as it encourages creation and learning. That inversion is the whole moral of the story. A copyright that does not, on net, produce more and better creation and more widely shared knowledge is not serving its purpose; it is just an enclosure. And this is the test the Statute of Anne hands us for every AI-content proposal, sharper than any argument about whether a diffusion model is an author: does this rule encourage learning and creation, or does it merely enclose them?
So here is the practical residue, and it is a way of reading, which you can apply to the next AI-copyright headline and to your own work.
When you meet a proposed rule for AI-generated content, run it through the three moves in order. First, does it correctly locate authorship in a human's creative contribution rather than in the machine or its operator? That part is mostly settled, and settled correctly; be suspicious of anyone reopening it, and pay attention instead to the genuinely hard assisted-middle cases like Zarya, where the question is how to credit the human selection and arrangement layered on top of machine generation. Second, and most important, ask whose coat the rule is wearing: trace who actually captures the value, and treat "we are protecting creators" as a claim to be verified, not accepted, because the first copyright law in history was a monopoly bill in exactly that disguise. Third, ask what the limits are, how the right expires and what feeds the public domain, because that is the move that served the public and the move the current debate keeps skipping.
And for your own AI-assisted work, the Zarya decision is a concrete instruction, not just a curiosity. The protectable thing is what you did, not what the model did: your selection, your arrangement, your editing, your judgment, the human creative contribution laid over the generation. Document that layer, because that is the part the law will recognize as yours, and understand that the raw generation is not yours and, under current law, is not anyone's. Build on that honestly rather than pretending the machine's output is your authorship, which the courts have now declined, all the way to the Supreme Court, to accept.
The Stationers won in 1710 by learning to state their interest in the language of the public good, and the law they helped create outlived their monopoly by centuries because, almost by accident, it also served that public good, through limits and a public domain they never wanted. Three hundred and fourteen years later we are writing the rules for the next copying machine. The Statute of Anne is not a relic to be updated for AI. It is a warning and a template, and its oldest question is still the only one that matters: not who owns what the machine made, but whether the rule we write around it will encourage learning, or simply fence it in.
The protectable thing is what you did, not what the model did. So keep the record of what you did.
The Zarya decision is a standing instruction for anyone shipping AI-assisted work: the law recognizes your selection, your arrangement, your editing, and your judgment layered over the generation, and not the raw output. That layer only counts if you can show it. Chain of Consciousness is exactly that record, a provenance trail of the human and agent decisions behind an artifact, so the contribution the courts will recognize as yours is documented rather than asserted after the fact. When ownership turns on where the human creative act actually was, the answer should not be a memory. It should be a log.
Hosted Chain of Consciousness · Read the Theory of Agent Trust
pip install chain-of-consciousness · npm install chain-of-consciousness
Or the full stack: pip install agent-trust-stack / npm install agent-trust-stack.