AI copyright law in short
US copyright law protects "original works of authorship" under 17 U.S.C. § 102(a), and courts and the Copyright Office read "authorship" to mean human authorship. That single requirement answers most questions about AI output. It does not answer the separate question of whether building an AI model infringes the copyrights in the works it was trained on, which is being fought case by case under the fair use doctrine.
The table sets out where each core question stands as of November 9, 2026.
| Question | Current answer | Key authority | Status |
|---|---|---|---|
| Can AI be an author? | No | Thaler v. Perlmutter (D.C. Cir. 2025) | Final; Supreme Court denied review March 2, 2026 |
| Can prompts alone make you the author of an AI image? | No, in the Copyright Office's view | Copyright Office Part 2 report (January 29, 2025) | Being tested in Allen v. Perlmutter (D. Colo.) |
| Is human selection, arrangement or editing of AI material protected? | Yes, to the extent of the human contribution | Zarya of the Dawn decision (February 21, 2023); Part 2 report | Settled Office practice |
| Is training on lawfully acquired books fair use? | Yes, in two district court rulings | Bartz v. Anthropic; Kadrey v. Meta (N.D. Cal., June 2025) | Not reviewed on appeal |
| Is training a competing non-generative tool fair use? | No | Thomson Reuters v. Ross (3d Cir. 2026) | ROSS says it will seek Supreme Court review |
| Who owns AI output? | No copyright owner for purely AI parts; contracts allocate other rights | Part 2 report; provider terms of use | Settled for copyright; contract terms vary |
For a running list of the lawsuits, see our AI lawsuit tracker.
Can AI generated content be copyrighted?
The human authorship requirement
The clearest authority is Thaler v. Perlmutter. Stephen Thaler sought to register an image, A Recent Entrance to Paradise, listing his AI system as the sole author and disclaiming any human creative input. The D.C. Circuit affirmed the refusal in March 2025, holding that the Copyright Act requires a human author. The Supreme Court denied certiorari on March 2, 2026 in No. 25-449, leaving that ruling in place. A cert denial sets no precedent of its own, but no federal appellate court has held otherwise. Our analysis of authorship after Thaler covers what the decision did not decide.
Because Thaler disclaimed any human role, the court did not say how much human involvement is enough. That is the live question, and the Copyright Office has answered it in guidance rather than regulation.
AI art and images
The Office's Part 2 report on copyrightability, published January 29, 2025, concludes that prompts alone do not give a user enough control over the output to make the user its author, even when many prompts are used. It also concludes that human expression that is perceptible in the output, such as a user's own drawing used as an input, can be protected, as can creative selection and arrangement of AI material and human modifications to it. The Office registers the human contribution and excludes the machine-generated material. The Office's registration decisions, including Zarya of the Dawn, are collected on its AI page.
AI music and code
The same rule applies to every category of work. The Part 2 report treats music, text, code and visual art under one human authorship standard: a song's human-written lyrics or melody can be protected while AI-generated audio is not, and the human-written portions of a program can be protected while generated code is not. No court had ruled on an AI music or code registration as of November 9, 2026, so these applications rest on the Office's general analysis.
Who owns the copyright in AI generated content?
Ownership follows authorship. Where a work contains no human authorship, there is no copyright and therefore no copyright owner: the purely AI-generated material is free for anyone to copy as far as copyright is concerned. Where a human contributes protectable expression, that person (or their employer, for a work made for hire) owns the copyright in that contribution only. The AI developer is not an author of a user's output under US law simply because it built the tool.
Contracts fill the gap. Many AI providers allocate rights in outputs through their terms of service. OpenAI's terms of use, for example, assign to the user whatever rights OpenAI has in the output, if any. An assignment cannot create a copyright that the law does not recognise, so these clauses matter mainly for trade secrets, contract claims and any human-authored material.
Other countries take different positions. The UK's Copyright, Designs and Patents Act 1988, § 9(3) treats the author of a computer-generated work with no human author as the person who made the arrangements necessary for its creation, a rule that has no US counterpart.
Is training AI on copyrighted work legal?
Training a model copies works, so the question is whether that copying is fair use under 17 U.S.C. § 107. Courts weigh the purpose and character of the use, the nature of the work, the amount used and the effect on the market for the original. The first merits rulings arrived in 2025 and 2026, and they point in different directions. Our fair use analysis covers the doctrine in depth.
Fair use rulings so far
In Bartz v. Anthropic, No. C 24-05417 WHA (N.D. Cal.), Judge William Alsup held on June 23, 2025 that training on books and scanning lawfully purchased print copies were fair use, but did not resolve the claim over books downloaded from pirate libraries on summary judgment. Two days later, in Kadrey v. Meta, No. 23-cv-03417-VC (N.D. Cal.), Judge Vince Chhabria granted Meta summary judgment on fair use, while stressing that the ruling rested on the record those plaintiffs built, particularly on market harm. Debevoise's comparison of the two rulings sets out where the judges' reasoning diverged.
Bartz then settled. Judge Araceli Martínez-Olguín granted final approval of the USD 1.5 billion class settlement on July 20, 2026, which pays roughly USD 3,000 per covered work, requires destruction of the pirated files and releases only past input claims, not claims over outputs.
The first appellate ruling went the other way on different facts. In Thomson Reuters v. Ross Intelligence, Judge Stephanos Bibas held in February 2025 that ROSS's use of Westlaw headnotes to train a competing legal research tool was not fair use. The Third Circuit affirmed in late September 2026 in an opinion initially filed under seal, according to LawSites, and ROSS has said it will ask the Supreme Court to review it. The ROSS system was not generative, which limits how far the ruling reaches.
Pending cases
Most generative AI training suits remain undecided on the merits, including The New York Times v. OpenAI, Andersen v. Stability AI and, in England, Getty Images v. Stability AI. Status for each is kept current on our case tracker rather than repeated here.
What has the Copyright Office said about AI?
The Copyright Office launched its AI initiative on March 16, 2023 and has published guidance and a multi-part report, all collected on its Copyright and Artificial Intelligence page.
- Registration guidance (March 16, 2023). The policy statement at 88 Fed. Reg. 16,190 requires applicants to disclose AI-generated material that is more than de minimis, claim only the human-authored parts, and correct earlier applications that omitted the disclosure.
- Part 1, Digital Replicas (July 31, 2024). The report recommended a federal law against unauthorised digital replicas of a person's voice or likeness.
- Part 2, Copyrightability (January 29, 2025). The report concluded existing law is adequate, prompts alone are insufficient, and human expression, selection, arrangement and modification are protectable.
- Part 3, Generative AI Training (pre-publication, May 9, 2025). The pre-publication report concluded that some training uses will be fair and others will not, depending on the facts. The Office says a final version will follow without substantive changes; it had not been posted as of November 9, 2026.
The reports set out the Office's views and its registration practice. They do not bind courts, which decide authorship and fair use questions for themselves.
Beyond copyright: patents and trademarks
Patents and inventorship
Patent law reaches the same result as copyright by a different route. The Federal Circuit held in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), that an inventor under the Patent Act must be a natural person. In November 2025 the USPTO rescinded its February 2024 guidance and issued revised inventorship guidance for AI-assisted inventions, applying the ordinary conception test to all inventions and confirming that an AI system cannot be named as an inventor. A human who conceives the invention using AI as a tool can still obtain a patent.
Trademarks
Trademark rights come from use of a mark in commerce to identify the source of goods or services, not from authorship. A logo or brand name generated with AI can therefore function as a trademark once a business uses it that way, even though the same image may carry no copyright. The practical risk runs the other direction: an AI tool may output a design that resembles an existing mark, and the user, not the tool, bears the infringement risk.
Creators should assume purely AI-generated output has no copyright owner; only human selection, arrangement and edits are protected, and the Supreme Court declined to revisit that rule on March 2, 2026. For AI developers, training on copyrighted work remains unsettled: Bartz and Kadrey found fair use in 2025, but the Third Circuit affirmed that training in Thomson Reuters v. Ross was not. The largest result so far is the $1.5 billion Bartz settlement.
What changed
- 2026-09-29: Third Circuit affirmed the ruling that ROSS's training on Westlaw headnotes was not fair use; opinion initially filed under seal.
- 2026-07-20: Court granted final approval of the USD 1.5 billion Bartz v. Anthropic settlement.
- 2026-03-02: Supreme Court denied certiorari in Thaler v. Perlmutter (No. 25-449).
- 2025-11-28: USPTO published revised inventorship guidance for AI-assisted inventions, rescinding its 2024 guidance.
- 2025-06-25: Kadrey v. Meta: summary judgment for Meta on fair use.
- 2025-06-23: Bartz v. Anthropic: training held fair use; pirated library claims left for trial.
- 2025-05-09: Copyright Office released the pre-publication Part 3 report on generative AI training.
- 2025-03-18: D.C. Circuit affirmed in Thaler v. Perlmutter that copyright requires a human author.
- 2025-01-29: Copyright Office published Part 2 of its AI report, on copyrightability.
Frequently asked questions
Can AI generated content be copyrighted?
Not the purely AI-generated parts. The Copyright Office registers only human authorship, so a work made entirely by an AI from prompts is not protected. Human-written text, human selection and arrangement of AI material, and meaningful human edits can be registered, with the AI material excluded from the claim.
Who owns the copyright in AI output?
Purely AI-generated output has no copyright owner in the US, because there is no human author. A person who adds protectable expression owns that contribution. Provider terms of service, such as OpenAI's, assign whatever rights the provider holds to the user, but cannot create a copyright the law does not recognise.
Is training AI on copyrighted work legal?
It depends on fair use, and courts are split. Two California district courts found training fair use in June 2025, while the Third Circuit affirmed in 2026 that ROSS's training of a competing legal research tool was not. Most generative AI training cases are still pending as of November 9, 2026.
What are the main AI copyright lawsuits?
The leading cases include Thaler v. Perlmutter on authorship, Bartz v. Anthropic and Kadrey v. Meta on training, Thomson Reuters v. Ross, The New York Times v. OpenAI, Andersen v. Stability AI and Getty Images v. Stability AI. Our case tracker keeps the status of each current.
Sources
- US Copyright Office: Copyright and Artificial Intelligence
- US Copyright Office: Part 2 report, Copyrightability (January 2025)
- US Copyright Office: Part 3 report, pre-publication (May 2025)
- Federal Register: AI registration guidance, 88 Fed. Reg. 16,190
- US Copyright Office: Zarya of the Dawn decision
- 17 U.S.C. 102 (Cornell LII)
- Mayer Brown: Supreme Court denies review in Thaler
- Debevoise: Anthropic and Meta decisions on fair use
- Authors Guild: final approval of the Anthropic settlement
- LawSites: Third Circuit rules for Thomson Reuters
- Federal Register: USPTO revised inventorship guidance
- legislation.gov.uk: CDPA 1988, section 9
This page is journalism and commentary, not legal advice. AI law changes quickly; check the primary sources above and consult qualified counsel for your situation.