Who Owns the Output?

Fair Use

The defense in 17 U.S.C. § 107 that permits some unlicensed use of a copyrighted work, judged by four factors: the purpose and character of the use, the nature of the work, the amount taken, and the effect on the market. Every AI training case turns on how a court weighs those factors against a use no one imagined when the statute was written.

Transformative Use

A use that adds a new purpose or meaning rather than substituting for the original, which weighs heavily in favor of fair use under the first factor. AI companies argue that turning millions of books into statistical weights is the most transformative use imaginable; authors answer that the output competes with the works that trained it.

Market Harm

The fourth factor, asking whether the use displaces the market for the original or for licenses to it. Courts have split over whether a licensing market for training data counts, and the Justice Department’s 2026 brief argues that it should not.

Training and Piracy

Two separate questions the Anthropic court kept apart: whether learning from a book is fair use, and whether downloading a pirated copy to learn from it is. The company won the first question and paid $1.5 billion on the second.

Human Authorship

The requirement that a copyrightable work originate with a human being. A work produced entirely by a machine cannot be registered, while a work that includes enough human creative contribution can be, with the machine-generated portions disclaimed.

Output Infringement

A claim that what a model produces, rather than what it was trained on, copies a protected work. It is the theory behind the news and lyrics cases, and it depends on showing that the output reproduces expression rather than facts or style.

Licensing Market

The deals AI companies now sign with publishers, labels, and news organizations for permission to train. Whether those deals are a courtesy or a legal necessity is what the fair use rulings will decide.

Part 1: Learn the Doctrine

Read § 107 and the Authors Guild explanation of the Anthropic ruling and settlement. Then use the Fair Use Tutor to work through the four factors with a training-data example until you can explain, in your own words, why the Anthropic court could find training fair and piracy not.

Part 2: Pick a Live Case and Argue Both Sides

Choose one pending case from a tracker in the Resources, and use the Docket Navigator to help you find its primary documents on CourtListener or the court’s site: the complaint, the answer or motion to dismiss, and any ruling on fair use or authorship. Read at least one filing from each side yourself. Then use the Two-Sided Advocate to draft the strongest 300-word argument for the plaintiff and the strongest 300-word argument for the defendant on the central copyright question, and revise each until every factual claim is traceable to a filing you read. Check any case the AI cites against Google Scholar before it stays in your brief.

Part 3: The Registration Exercise

Generate an image from a theme you supply with any AI image tool, then generate a second version in which you make a real creative contribution: a detailed composition you designed, hand edits, or a written work the image illustrates. Use the Registration Guide to walk through the Copyright Office’s guidance and decide what, if anything, in each version you could register and what you would have to disclaim. Write a short explanation, of about 200 words, addressed to a classmate who assumes they own everything an AI makes for them.

Part 4: Post Your Deliverables

Create a single new discussion thread containing the name and docket of your case with links to the filings you read, your two 300-word arguments, your registration analysis with both images, and a paragraph on which of the three questions in this activity, training, output, or ownership, you now think is hardest and why.

Part 5: Share Chat Link

Include one AI chat link with a 1–2 sentence explanation of what the conversation shows and why you chose to share it.

Fair Use Tutor

Act as a copyright law professor teaching a college student. Walk me through the four fair use factors in 17 U.S.C. § 107 one at a time, using the example of a company that trains a language model on published books. For each factor, explain what courts ask, what the AI company would argue, what the authors would argue, and how a court might weigh it. Then explain why a court could find the training itself fair use while finding that downloading pirated copies of the books was not. After each factor, ask me a question to check my understanding before moving on, and correct me where I am wrong. Name any case you rely on with enough detail that I can look it up, and tell me plainly if you are unsure whether a case exists or what it held.

Factor-by-factor with a comprehension check. Separating the factors keeps the analysis honest about which one does the work, and the instruction to name cases findably and flag uncertainty protects you from a hallucinated precedent in the one area where they are most common.

Docket Navigator

Act as a litigation paralegal who specializes in finding court records. I will name a pending AI copyright case. Help me locate its primary documents: the court and docket number, where the docket is available online (CourtListener, the court’s website, or PACER), and which filings I should read first to understand the central copyright question, such as the complaint, the motion to dismiss or for summary judgment, and any ruling. Give me search terms rather than guessing at document numbers, and tell me when you are not certain a document exists or where it is. Do not summarize the case for me; my task is to read the filings myself.

Finding, not summarizing. The prompt keeps the model in the role of a guide to the record and tells it to admit uncertainty, so you end up reading the actual filing instead of the model’s memory of it.

Two-Sided Advocate

Act as an appellate advocate who can argue either side of a copyright dispute with equal skill. I will describe a pending AI copyright case and paste key passages from the filings I have read. First, write the strongest 300-word argument for the plaintiff on the central copyright question, using only facts from the filings and authority you can name precisely. Then write the strongest 300-word argument for the defendant under the same rules. Do not tell me which side is right. After both arguments, list every case or statute you cited so I can verify each one before I rely on it.

Symmetry as a discipline. Requiring an equally strong argument for each side keeps the model from collapsing into whichever view is more common in its training data, and the citation list at the end is your verification checklist.

Registration Guide

Act as a copyright registration specialist. I will describe two images: one generated entirely by an AI tool from my text prompt, and one to which I added a real creative contribution, which I will describe. Walk me through the U.S. Copyright Office’s guidance on works containing AI-generated material and tell me, for each image, what if anything I could register, what I would have to disclaim, and why prompting alone does not count as authorship. Then help me explain the answer in plain language to a classmate who believes they own everything an AI makes for them. Cite the specific guidance you rely on so I can confirm it on the Copyright Office’s site.

Applying agency guidance to your own work. Using an image you made turns an abstract authorship rule into a decision about something you care about, and the citation requirement points you back to the Office’s own documents.