Practice: Who owns AI's work

Self-check

Six short questions. Answer each in your head before opening the collapsible. Active retrieval is where the learning sticks.

1. Why does copyright exist at all, in the view that drives United States law?

Show answer

Because of a bargain. Society hands a creator a limited, temporary head start over what they made, so for a while others need permission to copy it. That lets the creator earn from the work, which gives people a reason to keep creating. When the protection ends, the work enters the public domain, meaning anyone can use it freely. So copyright is a deal struck for a purpose: more creation, for everyone, over the long run. An older, natural-rights view, that what I make is mine and taking it feels like theft, carries less legal weight but shapes how people feel.

2. Is training an AI on copyrighted work fair use, and how do the recent rulings fit?

Show answer

It is a genuinely open question as of July 2026. Since 2024, real courts have ruled, but each is a narrow, fact-specific decision from a single trial court, and none has been ruled on by a higher court. One Delaware court found a startup’s copying not fair use; two California courts found a particular training use fair use, and both hedged hard, with one already argued on appeal. The rulings conflict and no higher court has settled the matter. So the honest summary holds: do not let anyone tell you it has been answered, in either direction.

3. What four things does a fair-use question turn on?

Show answer

The purpose of the use (a use that adds something genuinely new is transformative, which weighs toward fair use); the nature of the work (plain facts get thinner protection, creative work gets more); the amount used (a little weighs toward fair use, the whole thing weighs against it); and the effect on the market, which often decides, because it asks whether the new use competes with the original and eats into its income.

4. Can you copyright what an AI makes, and what is the split for a user?

Show answer

The bedrock rule is human authorship: United States copyright protects work made by a human being, because the bargain is meant to give people a reason to create and a machine needs no such reason. Purely AI-made work, with no human author, is settled as not copyrightable, but only for that clear situation. The practical split, under the Copyright Office’s registration practice, is this: the bare output of a prompt is probably not yours, because typing prompts alone does not steer the expression closely enough, but your own creative selection, arrangement, and editing can be protected. The more of you in the final piece, the more there is to protect.

5. What did the Anthropic settlement involve, and what can we not read into it?

Show answer

After its ruling, Anthropic reached a deal to settle the piracy part of the Bartz case, agreeing to pay at least 1.5 billion dollars and to destroy its pirated copies; this went before a court for review in 2026. What we cannot read into it: a settlement is a deal to end a case, not a court finding, and settling is not an admission of anything. We report it as a fact and draw no lesson from it about who was right.

6. Where is the line between copying and style, and what other law may still apply?

Show answer

Copying a specific protected work is barred; imitating a general style is not. You are allowed to write a song in the style of an artist you admire, but you cannot reproduce a particular song. A work built directly on a specific protected work is a derivative work and needs the owner’s permission; a work merely in a similar style is not. A separate body of law, the right of publicity, protects a person’s name, voice, and likeness, so a voice clone could run into that even when it does not break copyright. The course leaves open whether style should still be the line when a machine can imitate almost anyone at scale.

Try it yourself

Two exercises, and an optional third in Clawless. The first two you can do on paper.

Exercise 1: Sort the claims into the three buckets

Below are seven short statements you might meet in the news. For each, do two things. First, label which bucket it belongs in: a decided ruling with limits, a settlement with no merits inference, or a genuinely open question. Second, if the statement is worded too strongly for its bucket, rewrite it so it is honest. Some are stated fairly; some are not. Then open the key.

A court found a startup’s copying of a legal-research company’s material, used to build a rival tool, was not fair use.
Training on copyrighted work is legal now.
An AI company agreed to pay at least 1.5 billion dollars to settle a claim that it kept pirated copies of books.
The Supreme Court declined to hear the case about an image claimed to be made by a machine with no human author, which left the lower ruling in place.
An artist’s registration for a prize-winning image he made with an AI tool is still being fought in court.
Making a song in the general style of a famous artist always breaks copyright.
Two courts have now settled that training AI on copyrighted books is fair use.

Show answer key

Decided ruling, with limits. Worded fairly, as long as you carry the limits: this was one trial court, the tool was not a generative AI, the decision turned on its own facts, and it has since been argued on appeal with a decision still to come. A decided fact of record, not a nationwide rule.
The trap. There is no such settled rule. This is stated far too strongly, in the wrong direction, and no bucket makes it true. Honest rewrite: whether training AI on copyrighted work is fair use is a genuinely open question as of July 2026; individual rulings are narrow, fact-specific, and conflicting, and no higher court has settled it. Note the honest version does not flip to the opposite overclaim either: do not rewrite it as “training was ruled infringement.” Both directions are open.
Settlement, no merits inference. Worded fairly. Keep the hedge at “at least 1.5 billion dollars” and add the discipline: a settlement is not a court finding, and settling is not an admission of anything.
Decided ruling, with limits. Worded fairly. The limit is the whole point: that case was about a work claimed to have no human involvement at all. It did not decide the far more common AI-assisted situation.
Genuinely open. Worded fairly. As of July 2026 there is no final ruling; this is the AI-assisted line, exactly where the fight now sits.
Overstated. Imitating a general style generally does not break copyright; copying a specific protected work is what can infringe. Honest rewrite: making a song in a famous artist’s general style generally does not break copyright, though reproducing a specific song can, and a separate law like the right of publicity may still apply to a voice.
Overstated. Two trial courts found a particular training use fair use, and both hedged hard; the rulings are narrow and conflicting, one has already been argued on appeal, and no higher court has settled anything. Honest rewrite: two trial courts ruled a specific training use fair use in narrow, fact-specific decisions, so the general question stays open.

Every item maps to a claim in the lesson. The single most useful habit here is telling a ruling from a rule: one court deciding one case on its own facts is a data point, not settled law.

Exercise 2: What do I own?

This one is a reflection, not a ruling. Picture two versions of the same piece.

Version A: you type a single prompt into an AI image tool and keep the first picture it hands back, exactly as it came out.
Version B: you take that same tool, run many prompts, pick among dozens of results, then crop, recolor, rearrange elements, and paste in a caption you wrote yourself.

Now reason, using the Copyright Office’s registration practice as the lesson described it (its own practice for handling registrations today, not a final word from the courts). For Version A, how much of the piece is a product of your own creative choices, and how much is the raw output of a prompt? For Version B, name the specific human contributions you added, the selection, the arrangement, the edits, the words you wrote, and ask which of those a person could point to as their own creative work. You are not deciding a legal case or your own; you are practicing the split the lesson taught: the bare output of a prompt is probably not yours to own, while the creative work you do with it quite possibly is, and the more of you in the final piece, the more there is to protect.

Exercise 3 (optional): Talk it through in Clawless

This step runs in Clawless, the working environment we use across Clawdemy, and a plain conversation is all it needs. A model is not a lawyer, so do not ask it to rule on your own situation. Instead, use it to sharpen your own understanding:

Ask the model to explain the four fair-use factors back to you in plain words, then check its answer against the lesson: the purpose of the use, the nature of the work, the amount used, and the effect on the market.
Ask it to role-play a skeptical editor who keeps pushing you to say which bucket a claim belongs in. Feed it the statements from Exercise 1 and defend your bucket call out loud, in your own words. If the editor pushes you toward “it is decided” or “it is illegal,” hold the line: an open question stays open.

Keep it to plain conversation. The point is to rehearse the buckets and the factors until they are yours, not to get an answer about a real dispute.

Flashcards

Q. Why does copyright exist at all?

Because of a bargain. Society hands a creator a limited, temporary head start over what they made, so for a while others need permission to copy it. That lets the creator earn from the work, which gives people a reason to keep creating. The protection does not last forever; when it ends, the work enters the public domain, meaning anyone can use it freely. This incentive bargain is the reason that drives United States law.

Q. What are the four factors a fair-use question turns on?

The purpose of the use, where a use that adds something genuinely new is transformative and weighs toward fair use. The nature of the work, where plain facts get thinner protection than creative work. The amount used, where taking a little weighs toward fair use and taking the whole thing weighs against it. And the effect on the market, which often decides, because it asks whether the new use competes with the original and eats into its income.

Q. What are the three buckets for a legal claim about AI, with one example each?

A decided ruling with limits: a Delaware court found a startup’s copying, used to build a rival tool, was not fair use, though the tool was not a generative AI and the case has since been argued on appeal. A settlement with no merits inference: Anthropic agreed to pay at least 1.5 billion dollars to end a piracy claim, which is not a court finding. A genuinely open question: whether the New York Times case against OpenAI is infringement, still in discovery with no ruling as of July 2026.

Q. Is training an AI on copyrighted work fair use?

It is a genuinely open question as of July 2026. Since 2024, courts have ruled, but each is a narrow, fact-specific decision from a single trial court, none reviewed by a higher court, and they conflict: one found a startup’s copying not fair use, two found a particular training use fair use while hedging hard. So no higher court has settled the matter. Do not let anyone tell you it has been answered, in either direction, neither that training is fair use nor that it was ruled infringement.

Q. What is the human-authorship rule?

United States copyright protects work made by a human being. The reason runs back to the bargain: copyright exists to give people a reason to create, and a machine does not need that reason, so a work with no human author has nothing the deal is meant to protect. This is settled at the top for one clear situation, a work claimed to be made with no human involvement at all, but it does not decide the far more common AI-assisted case.

Q. Do you own the output of a prompt, and what part can you own?

The bare output of a prompt is probably not yours, because, under the Copyright Office’s registration practice, typing prompts alone does not steer the actual expression closely enough (the same prompt can produce many different pictures). But your own creative choices can be protected: if you select, arrange, and edit AI output in a creative way, or feed in your own writing that shows up in the result, that human part can be copyrighted even though the raw AI-made part cannot. The more of you in the final piece, the more there is to protect.

Q. Where is the line between copying and imitating a style?

Copying a specific protected work is barred; imitating a general style is not. You are allowed to write a song in the style of an artist you admire, but you cannot reproduce a particular song. A work built directly on a specific protected work is a derivative work and needs the original owner’s permission; a work merely in a similar style is not. So a fake in someone’s style probably does not break copyright, even though it borrows the artist’s sound.

Q. What did the Anthropic settlement involve, and what can we infer from it?

After its ruling, Anthropic reached a deal to settle the piracy part of the Bartz case, agreeing to pay at least 1.5 billion dollars and to destroy its pirated copies; this went before a court for review in 2026. What we cannot infer: a settlement is a deal to end a case, not a court finding, and settling is not an admission of anything. We report it as a fact and draw no lesson from it about who was right.

Q. What is the Copyright Office's registration practice on AI output?

Set out in a 2025 report, and framed as the Office’s own practice rather than a final word from the courts: typing prompts alone does not make you the author, because you are not steering the expression closely enough and the same prompt can yield many pictures. But a human’s creative selection, arrangement, and editing of AI output, or human-authored writing that shows up in the result, can be protected. The Office registered one comic book on exactly this basis: the arrangement and the human-written words were protected, the AI-made images on their own were not.

Q. Define intellectual property, copyright, and the public domain.

Intellectual property is creations of the mind, writing, music, art, inventions, ideas, that the law treats a little like property even though you cannot hold them. Copyright is the branch that covers creative expression, the actual words, images, and sounds a person makes. The public domain is the state a work enters when its copyright protection ends, which just means anyone can use it freely.

Q. Define infringement, fair use, and transformative.

Infringement means using someone’s protected work, without permission, in a way the law forbids. Fair use is the escape hatch: a doctrine in United States law that allows some use of copyrighted work without permission, so copyright does not choke off things like teaching, news, research, and commentary. Transformative describes a use that adds something genuinely new rather than just repackaging the original, and it weighs toward fair use.

Q. Define human authorship, derivative work, and the right of publicity.

Human authorship is the rule that United States copyright protects work made by a human being. A derivative work is a new work built directly on top of a specific protected work, and making one needs the original owner’s permission; a work merely in a similar style is not a derivative work. The right of publicity is a separate law, recognized in many places, that protects a person’s name, voice, and likeness from being used without consent, so a voice clone could run into it even when it does not break copyright.