Explainer
Fair use explained: the four factors, key cases and AI training
Fair use is the US copyright rule that some unlicensed uses, for purposes such as criticism, comment, news reporting, teaching and research, are not infringement.[1] Courts weigh four factors, from the purpose of the use to its effect on the market for the original, and decide each case on its facts; there is no safe percentage or word count.[2][3] In 2023 the Supreme Court held in the Warhol case that adding new meaning is not enough on its own, and in 2025 two federal trial judges found the AI training uses before them to be fair use on those records.[4][5][6]
This explainer describes how US courts apply fair use, citing the statute, Supreme Court opinions and court orders. It is general information, not legal advice; whether a particular use is fair depends on its facts. Positions in pending lawsuits are the parties’ claims, not findings.
Fair use is the main limit on the exclusive rights described in how copyright works. Section 107 of the Copyright Act says that fair use of a copyrighted work, for purposes such as criticism, comment, news reporting, teaching, scholarship or research, is not an infringement.[1]
The four factors
There is no checklist that makes a use fair. Judges look at four things together:[2]
- Why and how you are using it: for example, whether the use is commercial or for nonprofit teaching.
- What kind of work it is.
- How much you take, and how important that part is.
- Whether your use hurts the market for the original.
There is no safe amount to copy. Courts decide each case on its own facts.[3]
Section 107 directs courts to consider the purpose and character of the use, including whether it is commercial or for nonprofit educational purposes; the nature of the copyrighted work; the amount and substantiality of the portion used in relation to the work as a whole; and the effect of the use on the potential market for or value of the work.[2] That the work is unpublished does not by itself bar a finding of fair use.[7] The Copyright Office stresses that courts decide fair use case by case on a fact-specific inquiry, with no formula fixing a percentage or number of words that may safely be used.[3] It keeps a searchable Fair Use Index of court decisions, sorted by category and type of use.[8]
Transformative use: Campbell to Warhol
In Campbell v. Acuff-Rose (1994), about 2 Live Crew’s parody of “Oh, Pretty Woman”, the Supreme Court said the first factor asks whether a new work merely supersedes the original or is “transformative”, adding new expression, meaning or message. The more transformative it is, the less other factors such as commercialism matter.[9] The Court held that a parody’s commercial character is only one element to weigh.[10]
In Google v. Oracle (2021), the Court held that Google’s copying of roughly 11,500 lines of declaring code from the Java SE interface, to let programmers work on Android, was fair use as a matter of law.[11]
Andy Warhol Foundation v. Goldsmith (2023) returned to the first factor. The use at issue was the foundation’s commercial licensing to Condé Nast of a Warhol image of Prince based on a photograph, and the Court held that the first factor did not favor the foundation.[12] New expression, meaning or message may be relevant but is not decisive on its own. The question is whether the use has a further purpose or different character, weighed against considerations such as commercialism.[4]
Fair use and AI training
Two federal judges ruled in June 2025 on whether training AI models on copyrighted books is fair use; the wider litigation is followed in the AI copyright lawsuits tracker.[5][6]
- In Bartz v. Anthropic, Judge William Alsup granted Anthropic summary judgment on 23 June 2025 that training on the plaintiffs’ books was fair use, as was scanning purchased print books.[5] He denied it for pirated copies kept in a central library and set them for trial.[13] The case later settled; the class settlement created a $1.5 billion fund.[14]
- In Kadrey v. Meta, Judge Vince Chhabria granted Meta summary judgment on 25 June 2025, finding its training use fair on the record before him.[6] He said this did not mean the training was lawful, only that the authors had made the wrong arguments.[15]
The US Copyright Office has also weighed in. Its report on AI is being published in parts, and the pre-publication Part 3, on training, came out in May 2025.[16] It said various training uses are likely transformative, but that commercial use of vast troves of works to produce competing expressive content, especially through illegal access, goes beyond established fair use.[17]
Trademark law has a separate test for expressive uses of brands; see how trademarks work.
Read next
AI copyright lawsuits tracker: rulings, settlements and open cases
Major AI copyright cases from court records: NYT v. OpenAI, the $1.5bn Bartz v. Anthropic settlement, Kadrey v. Meta and Getty v. Stability AI, dated.
Questions readers ask
What are the four fair use factors?
Section 107 lists the purpose and character of the use, including whether it is commercial; the nature of the copyrighted work; the amount and substantiality of the portion used; and the effect of the use on the potential market for or value of the work.[2]
Is there a safe amount I can copy, such as 10% or a few lines?
No. The Copyright Office says there is no formula, such as a set percentage or number of words, that guarantees a use is fair, and courts decide each case on its specific facts.[3]
What does "transformative" mean?
In Campbell v. Acuff-Rose (1994), the Supreme Court described a transformative use as one that adds new expression, meaning or message rather than merely superseding the original. In the Warhol case (2023) it added that new meaning alone does not decide the first factor; the use must have a further purpose or different character, weighed against things like commercialism.[9][4]
Can a commercial use be fair use?
Yes. Campbell held that a parody's commercial character is only one element to weigh, and in Google v. Oracle (2021) the Court held that Google's copying of about 11,500 lines of Java API code for Android was fair use.[10][11]
Is training AI on copyrighted books fair use?
Courts are still deciding. In June 2025, two federal district judges found the training uses before them fair use, but one sent pirated library copies to trial and the other said his ruling did not make the training lawful in general. The Copyright Office's 2025 pre-publication report said various training uses are likely transformative but some commercial uses go beyond established fair use.[5][13][15][17]
Sources
· 17 claims from 7 publishers
Each numbered claim is a statement we checked against the sources listed with it. Tap a number in the text to see its sources in place.
- [1]
Section 107 of the Copyright Act says fair use of a copyrighted work, for purposes such as criticism, comment, news reporting, teaching, scholarship or research, is not an infringement. confirmedas of 2026-11-07
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107 (retrieved 2026-11-07)
- [2]
Section 107 lists four factors courts must consider - the purpose and character of the use (including whether it is commercial), the nature of the work, the amount and substantiality of the portion used, and the effect on the potential market for or value of the work. confirmedas of 2026-11-07
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107(1) (retrieved 2026-11-07)
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107 (retrieved 2026-11-07)
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107(2) (retrieved 2026-11-07)
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107(3) (retrieved 2026-11-07)
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107(4) (retrieved 2026-11-07)
- [3]
The Copyright Office says courts decide fair use case by case on the specific facts, and there is no formula, such as a set percentage or number of words, that guarantees a use is fair. confirmedas of 2026-11-07
- U.S. Copyright Office Fair Use Index · US Copyright Office · Primary source · About Fair Use (retrieved 2026-11-07)
- U.S. Copyright Office Fair Use Index · US Copyright Office · Primary source · About Fair Use (retrieved 2026-11-07)
- U.S. Copyright Office Fair Use Index · US Copyright Office · Primary source · About Fair Use (retrieved 2026-11-07)
- [4]
Warhol held that new expression, meaning or message is relevant but not enough on its own; the first factor asks whether the use has a further purpose or different character, weighed against considerations such as commercialism. confirmedas of 2026-11-07
- Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2023-05-18 · Syllabus (a) (retrieved 2026-11-07)
- Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2023-05-18 · Syllabus (a) (retrieved 2026-11-07)
- [5]
On 23 June 2025, Judge William Alsup granted Anthropic summary judgment that using the plaintiffs' books to train its models was fair use, and that converting purchased print books to digital form was also fair use. confirmedas of 2025-06-23
- Bartz v. Anthropic PBC, No. C 24-05417 WHA, Order on Fair Use (Dkt. 231) · US District Court for the Northern District of California (via CourtListener RECAP) · Primary source · 2025-06-23 · Order on Fair Use (Dkt. 231, case 3:24-cv-05417-WHA), Conclusion (retrieved 2026-11-07)
- [6]
On 25 June 2025, Judge Vince Chhabria granted Meta summary judgment in Kadrey v. Meta, a suit by 13 authors, finding that Meta's use of their books to train its Llama models was fair use on the record before him. confirmedas of 2025-06-25
- Federal judge sides with Meta in lawsuit over training AI models on copyrighted books · TechCrunch · Secondary source · 2025-06-25 (retrieved 2026-11-07)
- Judge reluctantly sides with Meta in Sarah Silverman, authors' AI copyright suit · Courthouse News Service · Secondary source · 2025-06-25 · Hand-saved copy (fetch-source blocked with 403) (retrieved 2026-11-07)
- Meta wins AI copyright case, but judge says others could bring lawsuits · CNBC · Secondary source · 2025-06-25 (retrieved 2026-11-07)
- [7]
The fact that a work is unpublished does not by itself rule out fair use. confirmedas of 2026-11-07
- 17 U.S.C. 107: Limitations on exclusive rights: Fair use · US Government Publishing Office (govinfo) · Primary source · 17 U.S.C. 107 (retrieved 2026-11-07)
- [8]
The Copyright Office keeps a searchable Fair Use Index of court decisions, sorted by category and type of use, to help the public see which uses courts have found fair or not fair. confirmedas of 2026-11-07
- U.S. Copyright Office Fair Use Index · US Copyright Office · Primary source (retrieved 2026-11-07)
- [9]
In Campbell v. Acuff-Rose (1994), a case about a 2 Live Crew parody, the Supreme Court said the first factor asks whether a new work is "transformative", adding new expression, meaning or message, and that the more transformative it is, the less other factors such as commercialism matter. confirmedas of 2026-11-07
- Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 1994-03-07 (retrieved 2026-11-07)
- Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 1994-03-07 (retrieved 2026-11-07)
- Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 1994-03-07 (retrieved 2026-11-07)
- [10]
Campbell held that a parody's commercial character is only one element to weigh, not a bar to fair use. confirmedas of 2026-11-07
- Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 1994-03-07 (retrieved 2026-11-07)
- [11]
In Google v. Oracle (2021), the Supreme Court held that Google's copying of roughly 11,500 lines of declaring code from the Java SE API for Android was fair use as a matter of law. confirmedas of 2026-11-07
- Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2021-04-05 · Syllabus (retrieved 2026-11-07)
- Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2021-04-05 · Syllabus (3) (retrieved 2026-11-07)
- Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2021-04-05 · Syllabus (retrieved 2026-11-07)
- Google LLC v. Oracle America, Inc., 593 U.S. 1 (2021) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2021-04-05 · Syllabus (retrieved 2026-11-07)
- [12]
In Andy Warhol Foundation v. Goldsmith (2023), the Supreme Court held that the first factor did not favor the foundation's commercial licensing of a Warhol image of Prince, based on a photograph, to Condé Nast for a magazine. confirmedas of 2026-11-07
- Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2023-05-18 · Syllabus, Held (retrieved 2026-11-07)
- Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2023-05-18 · Syllabus (retrieved 2026-11-07)
- Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, 598 U.S. 508 (2023) · Cornell Law School Legal Information Institute (opinion text) · Primary source · 2023-05-18 · Syllabus (retrieved 2026-11-07)
- [13]
The same order denied summary judgment on the pirated copies Anthropic kept in its central library and set them for trial. confirmedas of 2025-06-23
- Bartz v. Anthropic PBC, No. C 24-05417 WHA, Order on Fair Use (Dkt. 231) · US District Court for the Northern District of California (via CourtListener RECAP) · Primary source · 2025-06-23 · Order on Fair Use (Dkt. 231), Conclusion (retrieved 2026-11-07)
- [14]
The Bartz v. Anthropic class settlement created a non-reversionary fund of $1.5 billion covering 482,460 works, an estimated payment of about $3,000 per work. confirmedas of 2026-07-20
- Bartz v. Anthropic PBC, Order Granting Final Approval of Class Action Settlement (Dkt. 680) · US District Court for the Northern District of California (via CourtListener RECAP) · Primary source · 2026-07-20 · Order Granting Final Approval (Dkt. 680), page discussing adequacy of relief (retrieved 2026-11-07)
- Bartz v. Anthropic PBC, Order Granting Final Approval of Class Action Settlement (Dkt. 680) · US District Court for the Northern District of California (via CourtListener RECAP) · Primary source · 2026-07-20 · Order Granting Final Approval (Dkt. 680), objections section (retrieved 2026-11-07)
- Bartz v. Anthropic PBC, Order Granting Final Approval of Class Action Settlement (Dkt. 680) · US District Court for the Northern District of California (via CourtListener RECAP) · Primary source · 2026-07-20 · Order Granting Final Approval (Dkt. 680), fairness analysis (retrieved 2026-11-07)
- [15]
Judge Chhabria said the ruling did not mean Meta's training was lawful, only that these plaintiffs made the wrong arguments and failed to build a record for the right one. confirmedas of 2025-06-25
- Judge reluctantly sides with Meta in Sarah Silverman, authors' AI copyright suit · Courthouse News Service · Secondary source · 2025-06-25 · Hand-saved copy (fetch-source blocked with 403) (retrieved 2026-11-07)
- Federal judge sides with Meta in lawsuit over training AI models on copyrighted books · TechCrunch · Secondary source · 2025-06-25 (retrieved 2026-11-07)
- [16]
The US Copyright Office is issuing its Copyright and Artificial Intelligence report in parts; Part 1 on digital replicas was published on 31 July 2024, Part 2 on the copyrightability of AI outputs on 29 January 2025, and a pre-publication version of Part 3 on generative AI training on 9 May 2025. confirmedas of 2026-11-07
- Copyright and Artificial Intelligence · US Copyright Office · Primary source (retrieved 2026-11-07)
- Copyright and Artificial Intelligence · US Copyright Office · Primary source (retrieved 2026-11-07)
- [17]
In the Part 3 pre-publication report, the Copyright Office said that various AI training uses are likely transformative, but that commercial use of vast troves of copyrighted works to produce expressive content competing with them in existing markets, especially through illegal access, goes beyond established fair use boundaries. confirmedas of 2025-05-09
- Copyright and Artificial Intelligence, Part 3: Generative AI Training (pre-publication version) · US Copyright Office · Primary source · 2025-05-09 · Part 3 pre-publication version, Conclusion (retrieved 2026-11-07)
Revision history (1)
- Page created.
Created Nov 7, 2026. Last reviewed by an editor on Nov 8, 2026. Next scheduled review: Feb 7, 2027.
Cite this page
"Fair use explained: the four factors, key cases and AI training." ContentLora, updated Nov 7, 2026. https://contentlora.com/explain/fair-use-explained
Spotted an error? Suggest a correction or emailcorrections@contentlora.com.