Who owns AI-generated content?

"Who owns it?" sounds like one question, but it is really two. The first is about copyright: the legal right, granted by law, to stop others from copying a creative work. The second is about contract: what the company that runs the AI tool lets you do with what it produces. The answers can differ, and both matter if you plan to sell a book cover, publish an article or use an image in an ad.

This article explains what the official texts said as of this fall. It is an overview, not legal advice. Rules vary by country, including in Canada, where courts have not settled the question; for anything with real money at stake, ask a lawyer who practises in your jurisdiction.

What the US Copyright Office says

In March, the US Copyright Office published guidance on registering works that contain AI-generated material. Its starting point is old and firm: copyright protects only what a human author creates.

When a person types a prompt and an AI system produces a poem, a picture or a song, the Office says the machine, not the person, decides the expressive details: the words in each line, the colours, the composition. It compares a prompt to the instructions you might give an artist you commissioned. You say what you want; the artist decides how. So the generated material itself is not protected, and applicants must leave it out of their claim.

That does not mean any work touched by AI is unprotected. A person who selects and arranges AI material in a creative way, or changes it substantially, can claim copyright in that human contribution. The Office applied this in February to Zarya of the Dawn, a comic book with a human-written story and images made in Midjourney: the book as a whole was protected, the individual images were not. Using ordinary tools such as Photoshop to edit your own work does not count against you.

What the courts said in August

Stephen Thaler had asked the Office to register an image made by an AI system he built, which he called the Creativity Machine, listing the machine as the author. The Office refused, and he sued. On August 18, Judge Beryl A. Howell of the federal court in Washington, DC, sided with the Office, writing that "human authorship is a bedrock requirement of copyright." She added that harder questions lie ahead about how much human input is enough when artists use AI as a tool. Thaler said he would appeal.

What the provider's terms say

OpenAI's terms of use, last updated in March, say you own your input and that OpenAI assigns to you all its rights in the output, as long as you follow the terms. The company says this lets you use it for any purpose, including selling or publishing it. Three conditions come with that. Output may not be unique, so another user can receive the same or a similar result. You are responsible for making sure your use breaks no law. And you may not present AI output as written by a human when it was not.

Put the two layers together and you get a practical picture:

What do Napster, Cariou v. Prince and the remix movement teach us?

Each of these is a moment when new technology ran ahead of copyright law, and each points to a lesson for AI.

Napster was the file-sharing service that let people swap music files without the artists' permission. In 2001, in A&M Records v. Napster, a US court found it had helped users infringe copyright, and the service shut down. The lesson: building a business on other people's work without permission invites a legal reckoning. That echoes today's lawsuits by authors such as Sarah Silverman against OpenAI and Meta over the material used to train their models.

Cariou v. Prince concerned the artist Richard Prince, who used Patrick Cariou's photographs of Rastafarians in a series of altered collages. In 2013, the Second Circuit appeals court found that most of Prince's works were fair use because they transformed the originals. Courts weigh how much a new work adds. Yet "transformative" has limits: in May of this year, the US Supreme Court ruled against the Andy Warhol Foundation over Warhol's prints based on a photograph of the musician Prince, at least for the licensing use in that case.

RiP: A Remix Manifesto, Brett Gaylor's 2008 documentary, followed remix artists such as Gregg Gillis, known as Girl Talk, and asked whether copyright helps or blocks culture built from earlier culture. Robert Scales helped build the film's website and, in A Beginner's Guide to ChatGPT, draws a direct line from that debate to today's questions about AI.

Together they suggest three working habits. Know where your material comes from. Add real human work of your own, because that is what the law protects. And expect the rules to keep moving, as they did for music sharing and for remix art, so check again before you rely on what was true last year.