The central ethical dispute over AI-generated art is whether companies may train image generators on artists' work without asking or paying, and this review found no final court ruling in the United States or Canada that settles it for image models as of October 24, 2026. The U.S. Copyright Office wrote in a May 2025 report that some training uses will count as fair use and others will not, depending on which works were used, where they came from, and what the outputs do to markets. The best-known artist lawsuit in the United States is still in pretrial proceedings, and the Canadian case now before the Federal Court concerns authorship of an AI-assisted image, not training.
Key takeaways
- Three disputes run together in public debate: consent for training, credit for the finished image, and competition between generated images and working illustrators.
- The Copyright Office's Part 3 report, still labeled pre-publication, expects fair use to be decided case by case and treats copying from pirate sources to make competing content as unlikely to qualify.
- Andersen v. Stability AI, filed by artists in January 2023, has had claims of direct and induced infringement allowed to proceed, with a summary judgment hearing reported as moved to February 17, 2027.
- In Canada, the Federal Court is hearing CIPPIC's challenge to a registration that names an AI tool as co-author of an image, and the Music Publishers Canada intervention was filed on July 15, 2026.
Three disputes that get mixed together
Complaints about AI art tend to blur separate questions. The first is consent: image models learn from very large collections of existing pictures, and many artists never agreed to that use. The second is credit and authorship: who, if anyone, is the author of an image produced from a text prompt. The third is competition: a generated picture in an artist's recognizable manner can be sold in the same market as the artist's own work. Law treats these differently. Consent and competition fall under the rules on copying and fair use, while credit falls under rules on who counts as an author. A reader judging a claim can ask which of the three it concerns before judging whether it is right.
What the Copyright Office said about training
The Copyright Office has published its work on AI in parts, and its Copyright and Artificial Intelligence page lists Part 3, on generative AI training, as released in pre-publication form on May 9, 2025. The Office said it did not expect substantive changes in a final version, and the page still described it as pre-publication when checked on October 24, 2026.
The report concludes that several stages of building a generative model involve copying that implicates owners' exclusive rights, so the real question is whether fair use excuses it. It finds that many training uses serve a new purpose and so favor fair use, but says the outcome depends on what works were used, from what source, for what purpose, and with what controls on outputs. Using vast amounts of copyrighted work to produce expressive content that competes with it in existing markets, especially through illegal access, goes beyond established fair use limits in the Office's view. It also discusses market dilution, the idea that a flood of competing outputs harms creators even when no single output copies one work, as a factor-four consideration. On remedies, the Office judged government intervention premature, pointing to growing voluntary licensing and suggesting extended collective licensing if gaps persist.
Where artists' lawsuits stand
In Andersen v. Stability AI, illustrators sued Stability AI, Midjourney, and DeviantArt in federal court in the Northern District of California in January 2023. A commentary from the NYU Journal of Intellectual Property and Entertainment Law reports that Judge William Orrick allowed copyright claims to proceed in August 2024, including direct infringement and induced infringement, while dismissing other claims. A court-tracking blog reported on February 3, 2026 that the summary judgment hearing was moved from November 4, 2026 to February 17, 2027 at the plaintiff's request. A trial date has been reported inconsistently, so the docket is the authority.
Two June 2025 district court decisions involving book authors, Bartz v. Anthropic and Kadrey v. Meta, found that training on books could qualify as fair use on the records before the courts. They concern text, come from trial-level judges, and do not bind other courts, so they offer hints more than answers for image models.
Canada: authorship before training
The Samuelson-Glushko Canadian Internet Policy and Public Interest Clinic, known as CIPPIC, asked the Federal Court in file T-1717-24 to remove or correct a copyright registration for an image called SURYAST that lists its creator, Ankit Sahni, and the RAGHAV AI painting tool as co-authors. In a July 31, 2025 post, CIPPIC states that it has standing as an interested person, that the work has no human author because the user supplied only an idea, and that originality requires skill and judgment that an AI system cannot exercise.
The Federal Court allowed Music Publishers Canada to intervene in June 2026, and the group filed submissions on July 15, 2026. It argues that only a human can be an author, that a person can still receive protection for work made with AI help, and that courts should look case by case at the human's skill and judgment. No decision in the matter was found, and no hearing date was published in the sources reviewed. This review found no Canadian court ruling on whether training an image model on artists' work infringes copyright.
Status summary
| Question | United States | Canada |
|---|---|---|
| Is training on artists' work lawful? | Unsettled; Copyright Office Part 3 says case by case; no final ruling on image models found | No ruling found |
| Main artist case | Andersen v. Stability AI, pretrial | None found on training |
| Can an AI tool be a listed author? | Addressed in the Office's separate copyrightability report | Before the Federal Court in T-1717-24 |
| Is licensing available? | Office reports licensing is emerging but uneven by sector | Not assessed in the sources reviewed |
This article is general information, not legal advice. Court schedules and agency reports change, and anyone facing a specific dispute or contract question should consult a qualified lawyer or the primary documents.
FAQ
Is it legal to train an AI image model on an artist's work without permission?
It has not been settled by a final ruling for image models in either country, according to this review. The Copyright Office expects some uses to be fair and others not.
Has any artist won a case against an AI image company?
No judgment for artists was found. Andersen v. Stability AI has survived early challenges but has not reached summary judgment or trial.
Can an AI tool be listed as an author in Canada?
That is the question in the pending CIPPIC application. The registration exists, and the Federal Court has not issued a decision in the sources reviewed.
Why does market dilution matter?
The Copyright Office treats it as a way harm can occur when many competing outputs reduce demand for human work without copying any single piece.
Related reading
More in AI Policy and Ethics.
Last verified: October 2026
Sources
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version), U.S. Copyright Office, May 2025
- Copyright and Artificial Intelligence overview page, U.S. Copyright Office
- CIPPIC files Memorandum of Fact and Law in CIPPIC v Sahni, CIPPIC, July 31, 2025
- Music Publishers Canada intervenes in Federal Court case on AI and copyright, Music Publishers Canada, July 2026
- Andersen v. Stability AI: the landmark case unpacking the copyright risks of AI image generators, NYU Journal of Intellectual Property and Entertainment Law, 2024
- Sarah Andersen's copyright lawsuit gets pushed back again, ChatGPT Is Eating the World, February 3, 2026
- Federal Courts Find Fair Use in AI Training: Key Takeaways from Kadrey v. Meta and Bartz v. Anthropic, Jackson Walker, 2025