In the United States, material generated entirely by AI is not protected by copyright, but a person's own creative contribution to a work made with AI can be. The U.S. Copyright Office set out that position on January 29, 2025, and the Supreme Court declined on March 2, 2026 to review the one appellate ruling requiring a human author. Canada has no court decision on AI authorship yet, and whether training a model on copyrighted works is lawful remains unsettled in both countries.
Key takeaways
- The Copyright Office concluded that, with current technology, prompts alone do not give enough control over the output to make a user its author.
- A federal appeals court held on March 18, 2025 that the Copyright Act requires a human author, and the Supreme Court denied review on March 2, 2026.
- Canadian federal policy papers say case law points to a human exercising skill and judgment, but a Federal Court challenge over a registration naming an AI as co-author is still pending.
- On training, the Third Circuit ruled on September 29, 2026 against a non-generative legal-research tool and expressly left generative AI open.
- Whether a given AI-assisted work is protected depends on the facts of each work, so records of the human contribution matter.
What the U.S. Copyright Office decided on authorship
The Office's report, Copyright and Artificial Intelligence, Part 2: Copyrightability, states that questions of copyrightability can be resolved under existing law, without new legislation. Using AI tools to assist human creativity does not affect protection for the result. Copyright does not extend to purely AI-generated material, or to material where human control over the expressive elements is insufficient.
The report treats prompts as instructions, not authorship, concluding that on the technology generally available, prompts alone do not provide sufficient control. A person can still hold copyright in expression of their own that is perceptible in an AI output, in the creative selection, coordination, or arrangement of material, and in creative modifications. Each case is assessed on its facts, and the Office found no case for a new type of right for AI output. Applicants who submit work with more than a minimal amount of AI-generated material are expected to disclose it and describe the human contribution.
The Thaler case
Stephen Thaler listed his generative system, the Creativity Machine, as sole author of an image titled "A Recent Entrance to Paradise" and named himself only as owner. The Copyright Office refused registration, and the U.S. Court of Appeals for the D.C. Circuit affirmed on March 18, 2025 in No. 23-5233. It held that the Copyright Act of 1976 requires all eligible work to be authored in the first instance by a human being. The court did not decide whether the Constitution demands the same, and it did not reach Thaler's argument that he was the author by operating the system, because that point had been waived before the agency. The Supreme Court's docket for No. 25-449 shows the petition for review was denied on March 2, 2026. A denial is not a ruling on the merits.
Where Canada stands
The federal consultation on copyright and generative AI ran from October 12, 2023 to January 15, 2024. Its summary report on the responses notes that Canadian case law suggests authorship must be attributed to a human who exercises skill and judgment. Most participants wanted human authorship to stay central. The government said it would examine whether legislative or policy action could support substantial human authorship, and no bill has followed in the sources reviewed.
The open question sits in the Federal Court. The Canadian Intellectual Property Office issued registration 1188619 on December 1, 2021 for the work "Suryast," naming Ankit Sahni and the RAGHAV AI app as co-authors, through an automated online system that did not verify the application's particulars. The Samuelson-Glushko Canadian Internet Policy and Public Interest Clinic (CIPPIC) filed court file T-1717-24 on July 8, 2024 to have it expunged or corrected. Music Publishers Canada filed as an intervener on July 15, 2026, and no decision had been reported by then.
Training on copyrighted works
Authorship is one question. The other is whether copying works to train a model is infringement or fair use. In Thomson Reuters v. ROSS Intelligence, decided on September 29, 2026, the Third Circuit held that Westlaw headnotes were copyrightable and that ROSS's use of them to build a competing legal search tool was not fair use. The court stressed that ROSS's system was not generative and returned only existing passages. It cited two 2025 California district court decisions, Bartz v. Anthropic and Kadrey v. Meta, as addressing whether generative training serves a new purpose, and said the ROSS facts pointed the other way.
In Bartz, the Authors Guild reports that final approval of a USD 1.5 billion class settlement came on July 20, 2026. The release covers claims about past acquisition and copying through August 25, 2025, not claims over outputs or future conduct. The Copyright Office's Part 3 report on generative AI training was released in pre-publication form in May 2025, and the Office's AI page still lists the final version as forthcoming. In Canada, the Ontario Superior Court held on November 7, 2025 that a suit by news publishers against OpenAI may proceed in Ontario for six of ten defendants. That ruling was on jurisdiction and did not address the merits.
United States and Canada side by side
| Question | United States | Canada |
|---|---|---|
| Is fully AI-generated output protected? | No, per the Copyright Office and the D.C. Circuit | No court ruling; government report says case law points to a human author |
| Are human edits or selection protected? | Yes, for the human contribution | Not tested in court |
| Is training lawful? | Unsettled; one appellate ruling against a non-generative tool | No merits ruling; jurisdiction ruling in an OpenAI case |
| New legislation? | The Office saw no need for new AI-specific rights | Options under review, no bill in the sources reviewed |
A record-keeping checklist for creators
- Save drafts, sketches, and edit history that show human choices.
- Note which parts were generated and which were written, drawn, or composed by a person.
- Prepare to disclose AI-generated material if registering in the United States.
- Read the terms of the AI tool for ownership and permitted use.
- Treat training and output questions as separate, since a court may answer one without the other.
This article is general information and not legal advice. Court cases and policy papers change, and anyone with a specific copyright question should consult a qualified lawyer or the original texts.
FAQ
Can a work made with ChatGPT or an image generator be copyrighted?
Only in part. The Copyright Office says protection can cover a person's own expression, selection, arrangement, or creative changes, but not material produced by the system alone.
Did the Supreme Court rule that AI cannot be an author?
No. It denied review on March 2, 2026, which left the D.C. Circuit's March 18, 2025 decision in place without addressing it.
Does Canada allow copyright for AI-generated work?
No court has decided. Government material says authorship has been attributed to a human who exercises skill and judgment, and a Federal Court application over a registration listing an AI as co-author is pending.
Is training an AI on copyrighted books legal?
That has no single answer yet. The Third Circuit's September 29, 2026 ruling went against a non-generative tool, and it left the generative question open.
Related reading
- ethical questions around AI-generated art
- AI ethical dilemmas in practice
- how AI text generators work
- AI and its effects on society
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Copyright and Artificial Intelligence, Part 2: Copyrightability, U.S. Copyright Office, January 29, 2025
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version), U.S. Copyright Office, May 2025
- Thaler v. Perlmutter, No. 23-5233, U.S. Court of Appeals for the D.C. Circuit, March 18, 2025
- Docket No. 25-449, Supreme Court of the United States, March 2, 2026
- Thomson Reuters v. ROSS Intelligence Inc., No. 25-2153, U.S. Court of Appeals for the Third Circuit, September 29, 2026
- Court Grants Final Approval of USD 1.5 Billion Anthropic Copyright Settlement, The Authors Guild, July 2026
- Consultation on Copyright in the Age of Generative Artificial Intelligence: summary of responses, Innovation, Science and Economic Development Canada, 2025
- CIPPIC v Sahni, Federal Court file T-1717-24, CIPPIC case page
- Music Publishers Canada intervention notice, July 22, 2026
- Jurisdictional Ruling Opens Door for Canadian Copyright Claims Against AI Companies, Carters Professional Corporation, 2025