Fair Use

Fair use is a limitation on copyright that permits certain uses of protected works without the rights holder's permission or payment. Its best-known form is the open-ended doctrine in United States law, codified in Section 107 of the Copyright Act of 1976, which asks courts to weigh several factors in each case. It operates within copyright and intellectual property law, and it reflects a balance between authors' exclusive rights and the public interest in criticism, education, research, and the circulation of information, a concern closely tied to freedom of expression.

The doctrine matters because it shapes how books, films, music, and journalism are quoted and reused, how libraries and classrooms operate, and how search engines and other digital services handle copyrighted material. Its reach has grown into new areas, including the training of large language models on copyrighted text, where courts and legislatures are still working out the rules. Because fair use is a standard rather than a fixed list of permitted acts, its meaning has changed through case law and remains contested.

This entry first explains the legal character of fair use and its historical origins, then sets out the four statutory factors and the key case law on transformative use, including the Supreme Court's 2023 decision in Warhol. It then compares fair dealing and other national models, international norms under the WTO and Berne frameworks, and applications in education, libraries, digital technology, and AI. The final section covers criticism of the doctrine and the open questions that remain.

Definition and legal character

Fair use is best understood as a restriction on what copyright owners can demand, not as a freestanding entitlement that users carry with them. In United States law it is codified in 17 U.S.C. § 107, which sits among the limitations on the exclusive rights that copyright grants. A statutory exception to the copyright holder's bundle of exclusive rights, it permits unlicensed use where the balance of several factors favors the user. A rightsholder therefore has no power to block a use that qualifies, because the owner of a copyright cannot prevent others from making a fair use of the owner's work.

The doctrine differs from three neighboring mechanisms. A license is a grant the owner chooses to make, usually in exchange for payment, while fair use operates without the copyright owner's permission and without compensating the owner in certain circumstances. A work in the public domain is not restricted by copyright at all, so no defense is needed. Specific statutory exceptions, such as the provisions on library copying or face-to-face classroom performance, are narrow, closed categories defined in advance. Fair use is open-ended by comparison: it is a general standard applied to the facts of each case rather than a list of predetermined permissions.

The statute names several purposes, including criticism, comment, news reporting, teaching, scholarship, or research. These names are the statute's illustrations, and the list is not exhaustive. Naming a purpose opens the inquiry rather than closing it. Those examples do not establish automatic categories of fair use. A use that fits a named purpose may still fail the analysis, and an unnamed purpose may still qualify. The open wording lets courts respond to new media and new practices without amending the text.

Fair use is also an affirmative defense. The party accused of infringement is the one who must justify the copying. Once a copyright holder establishes valid ownership and copying, the burden moves to the user to justify the use. The Supreme Court has long treated the doctrine this way. In Campbell v. Acuff-Rose Music (1994), the Court confirmed this framework. In practice, this allocation means that a defendant who cannot produce evidence supporting the four factors may lose even where the outcome is uncertain on the merits.

The doctrine has no bright-line rules. There is no numerical safe harbor; on quantity, there are no bright lines, but the higher the percentage of the original work used, the more likely this factor is to weigh against fair use. Size alone is not decisive either: even if the percentage is fairly small, if the material used is central to the work, this factor may weigh against fair use. The four statutory considerations are the purpose and character of the use, the nature of the copyrighted work, the amount and substantiality of what was taken, and the effect on the market for the original. None is dispositive on its own, and there is no equation or formula for determining whether a use is fair. Judgment is exercised across the factors together, which is why outcomes depend heavily on the specific facts of each use.

Because the doctrine is flexible by design, it also serves a constitutional function in the American system. Harvard's copyright guidance describes it as one of the features that reconciles the copyright statute with the First Amendment, alongside the distinction between ideas and their expression. This framing explains why fair use is treated as a built-in safeguard for free expression rather than merely a technical exception to copyright enforcement.

Origins and codification

Roots before 1841

Accounts of fair use usually begin with Folsom v. Marsh (1841), but the doctrine has older roots. Matthew Sag argues that the American doctrine continues a long line of English "fair abridgment" cases that date back to the statutory copyright regime created by the Statute of Anne in 1710 (the fair use doctrine is better understood as the continuation of a long line of English fair abridgment cases, dating back to the beginning of statutory copyright law in 1710). The clearest early statement came in 1741, when Lord Hardwicke, in Gyles v. Wilcox, held that a book merely shortened in form was an evasion of the statute and not a true abridgment, so a distinction between legitimate condensation and disguised copying was already in place.

Folsom v. Marsh and the abridgment idea

Folsom v. Marsh arose from a two-volume biography of George Washington that reproduced many of his letters from a much larger edition (a complaint about a 2 volume biography of George Washington making use of his papers and letters which had been extracted (copied) from a much larger 12 volume work). Justice Joseph Story, sitting on circuit in Massachusetts, did not treat every derivative use as infringing. He asked whether the defendant's work was a "fair and bona fide" derivative use, and he weighed the nature and purpose of the selections, the quantity and value of the material taken, and the extent to which the new work might harm the sales of the original. Story accepted that some uses could be fair but held that this one was not, and the plaintiff won. The opinion is therefore an early model of the factor-based reasoning that later became the statute, though for most of the next century courts cited it chiefly on the question of infringement rather than as a fair use precedent.

Common-law development in the twentieth century

Through the 1909 Copyright Act and well into the 1970s, the doctrine lived almost entirely in case law. The United States had not codified any exceptions to copyright prior to the 1976 Act. Courts found plenty of room for exceptions, including some—such as those for libraries and for teaching—which in the 1976 Act would be codified separately, and under the 1909 Act each of these was treated as a form of fair use. Congress prepared for revision by commissioning a 1958 Copyright Office study by Alan Latman, which reviewed the legislative history of the 1909 Act and the case law since Folsom.

Codification in 1976

Section 107 of the Copyright Act of 1976 gave the doctrine statutory form. It lists examples of qualifying purposes, namely "criticism, comment, news reporting, teaching (including multiple copies for classroom use), scholarship, or research", and sets out four factors for courts to weigh. The Act's definitions treat the words "including" and "such as" as illustrative rather than limiting, so the list of purposes and factors is open-ended. Congress described its aim as clarifying existing judge-made law, not replacing it, and the 1976 text was meant to preserve courts' ability to decide case by case. Section 107 also states that the fact that a work is unpublished shall not itself bar a finding of fair use. Commentators note that the first decade of judicial interpretation after 1976 was generally cautious and even conservative.

Harper & Row (1985)

The most consequential pre-2000 ruling was Harper & Row v. Nation Enterprises (1985). The Nation published roughly 300 words from former President Gerald Ford's unpublished memoir just before Time planned its own serialization, which had been contracted by the publisher. The district court found infringement, and the Second Circuit reversed on fair use grounds. The Supreme Court reinstated liability. It treated the unpublished character of the manuscript as "a key, though not necessarily determinative, factor" tending to negate a defense of fair use, and it stressed that the commercial motive and the loss of a first-publication right weighed against the defendant. Later courts read the decision as declining to create a categorical bar on fair use of unpublished works, which left the case as a strong signal rather than a rule.

The four statutory factors

Section 107 of Title 17 of the United States Code names four considerations that a court must weigh when it decides whether a use is fair. The list is not exhaustive: the statute says the factors "shall include" these items, and courts may consider other matters as well While it "leav[es] open the possibility that other factors may bear on the question, the statute identifies none." That is, courts are entitled to consider other factors in addition to the four statutory factors. In practice, the four factors form the working structure of nearly every American fair use analysis, and legislators who drafted them in 1976 intended them to restate and clarify earlier judge-made law rather than replace it.

Factor one: purpose and character of the use

The first factor asks why the defendant copied and what the copy does. The statute itself singles out whether a use is of a commercial nature or is for nonprofit educational purposes. The drafting committee made clear that this is not a rule that educational uses are automatically protected: the commercial or non-profit character of an activity, while not conclusive with respect to fair use, can and should be weighed along with other factors.

The factor also asks whether the use is transformative, meaning whether it serves a different function or adds something to the original. Commerciality and transformation interact. In Harper & Row v. Nation Enterprises (1985), the magazine's article was classed as news reporting, yet the Court still weighed against it because it was published for profit, meant to supplant other news stories, and built on a knowing exploitation of a purloined manuscript. Although The Nation's use of Ford's work was news, it also was for profit, was intended to supplant other news stories, and was based on knowing exploitation of a purloined work. In Andy Warhol Foundation v. Goldsmith (2023), the Court held that a licensed silkscreen based on a photograph did not favor fair use, because the original and the copying use shared similar commercial purposes—portraits of Prince used in magazines to depict the musician. The majority treated new expression as insufficient by itself, a point discussed in the section on the Warhol decision.

Factor two: nature of the copyrighted work

The second factor looks at the work being copied. Courts have long recognized that the law generally recognizes a greater need to disseminate factual works than works of fiction or fantasy. Creative works such as novels, paintings, music, and photographs therefore usually receive stronger protection under this factor than factual compilations or reports.

Publication status is another key element. In Harper & Row, the Court held that the unpublished nature of a work is a key, though not necessarily determinative, factor tending to negate a defense of fair use. The Court added that under ordinary circumstances, the author's right to control the first public appearance of his undisseminated expression will outweigh a claim of fair use. The statute itself, however, provides that unpublished status does not by itself bar a finding of fair use. In lower courts, factor two is often given little weight when the use is strongly transformative, and it may be treated as neutral in such cases.

Factor three: amount and substantiality

The third factor asks how much was taken, measured against the work as a whole, but it is not a simple percentage test. In Harper & Row, the verbatim passages were a small share of the book, and the Court acknowledged that the verbatim excerpts constituted only 300 to 400 words of the manuscript, this use was considered qualitatively significant. The Court found that The Nation had appropriated "the heart" of the manuscript, focusing on the most expressive and valuable parts of Ford's narrative. The Court criticized the lower court for finding the taking "infinitesimal," [and] accord[ing] too little weight to the qualitative importance of the quoted passages of original expression.

The factor also considers whether the amount taken is reasonable in relation to the purpose. A defendant cannot justify copying by pointing to how much more it could have taken without doing so. Courts therefore examine both the quantity and the quality of what was copied, and a small excerpt may still weigh heavily against fair use if it captures the most distinctive part of the original.

Factor four: effect on the market or value

The fourth factor asks whether the use harms the market for the original, or the value of the original. In Harper & Row, the unauthorized excerpts disrupted a pending licensing arrangement. This unauthorized use "caused the Time agreement to be aborted, and thus diminished the value of the copyright." The Court also considered the impact on the first serial rights market, since prepublication excerpts could erode the author's ability to sell the work in that form.

In Warhol, the Court of Appeals had found that the licensing encroached on Goldsmith's protected market to license her photograph "to publications for editorial purposes and to other artists to create derivative works". This illustrates that factor four covers not only direct sales of the original but also licensing markets and derivative markets that the owner could reasonably exploit. Because factor four concerns substitution, a use that does not compete with the original in any market usually weighs less heavily against the defendant.

Balancing the factors

The statute does not assign weights or an order of analysis. It is important to note that every factor will play some role in the court's decision-making process, but the interaction between them is not always clear. A court may find that one factor is decisive while the others are neutral or cut the other way. The 2023 Warhol litigation shows how this works in practice: the Supreme Court limited its review to the first factor and agreed with the lower court on that issue alone, and it left the other three factors to the lower courts. On that narrow issue, and limited to the challenged use, the Court agrees with the Second Circuit: The first factor favors Goldsmith, not AWF. Because the four factors are weighed together rather than applied in sequence, a finding on one of them does not automatically resolve the case.

Transformative use in case law

Campbell v. Acuff-Rose Music (1994): parody as commentary

The Supreme Court's first major treatment of transformation in this context arose from a dispute over a 1964 Roy Orbison song. In Campbell v. Acuff-Rose Music, Inc., 510 U.S. 569 (1994), the Supreme Court ruled that the rap group 2 Live Crew did not violate copyright law with the song "Pretty Woman," a parody of the previously released "Oh, Pretty Woman." The Sixth Circuit had reasoned that the commercial character of the parody made it presumptively unfair. The Court unanimously rejected that presumption. Justice Souter's opinion held that a parody's commercial character is only one element to be considered in a fair use enquiry. The Court also held that the lower court had not given enough weight to the nature of parody when it measured how much of the original had been taken.

The reasoning gave lasting shape to the link between transformation and commentary. A parody must borrow enough of its target to be recognizable, because its critical point depends on the original. Borrowing for that purpose differs from borrowing to replace the work or to satisfy the same demand for it. The Court also found no proof that the parody had harmed a market for licensed rap derivatives; the opinion observed that artists ask for criticism, but only want praise, so a critical parody was unlikely to substitute for a licensed derivative. The case therefore set out a framework of case-by-case weighing rather than a rule that commercial use is presumptively unfair, and it placed commentary on the original at the center of the analysis.

Google LLC v. Oracle America (2021): functional reuse of software interfaces

Decided on April 5, 2021, the Supreme Court's ruling in Google LLC v. Oracle America resolved a dispute that had run for roughly a decade. By a vote of 6-2, the Supreme Court ruled for Google, relying on the fair use doctrine to conclude that Google's copying did not infringe Oracle's copyright in Java. The copied material was the "declaring code" of the Java SE application programming interface, about 11,500 lines, which Google reused in the Android platform. The Federal Circuit had held that the use was not fair as a matter of law, and the Supreme Court reversed that judgment.

Justice Breyer's majority opinion treated the purpose of the copying as functional. The Court gave weight to the fact that Google copied no more than the necessary portions of the Sun Java API in order to "create a new platform," the Android platform. The Court also found that the nature of a functional interface weighed toward fair use, because such code is a building block for task organization rather than expressive creative work. The holding was described as narrow: the Court's holding was relatively narrow, concluding that Google's copying of certain code from the Java API—what the Court characterized as "reimplementation of a user interface"—was a fair use under the "case-by-case" balancing of the statutory fair use factors. Justice Thomas, joined by Justice Alito, dissented, objecting both to the Court's decision to assume copyrightability and to its fair use conclusion. The decision is therefore best understood as extending transformation to technical reuse, not as a general license for copying software.

Authors Guild v. Google (2015): scanning books for search

On October 16, 2015, the Second Circuit unanimously affirmed the district court's ruling in the Google Books litigation. The project involved digitizing and creating a searchable database of 20 million books, including books protected by copyright. Google also displayed short excerpts in response to search queries, which the court treated as part of the same analysis. The Second Circuit affirmed, finding the project to be a highly transformative use within the scope of the fair use doctrine.

The court's reasoning rested on what the copies did rather than on what they displayed as literary content. Google's copy served to let the search function reveal limited information about each book, and the court stressed that the search function was making available information about Plaintiffs' books without providing the public with a substantial substitute for matter protected by the Plaintiffs' copyright interests. The court also drew on Campbell's language describing transformative works as lying at the heart of the doctrine's breathing space within the confines of copyright. The decision became a reference point for large-scale data uses, although its reliance on limited snippets and the absence of substitution was central to the result.

How "transformative" broadened

In Campbell, the term was tied to adding new expression, meaning, or message to the original. Over the following decades, courts applied it to uses whose purpose differed from the original's, including information retrieval, indexing, and technical functions. The Google Books and Google v. Oracle decisions show two directions of this expansion: the first treats a copy as transformative when it creates a tool that makes information about a work accessible, and the second treats reuse as transformative when it serves a different functional end in a new computing platform. Both move the inquiry from the content a reader or viewer perceives toward the function the copying serves. This expansion made transformation a central, and contested, part of factor one, a development that the later Warhol decision in 2023 sought to narrow, as discussed below.

The Warhol decision and limits on transformation

The dispute began with a 1981 photograph of the musician Prince taken by the rock photographer Lynn Goldsmith. Artist Andy Warhol created a series of silkscreen prints and pencil illustrations based on that copyrighted photograph. The first print in the series, an orange-hued image, ran in Vanity Fair, and Warhol later made fifteen more works from the same photograph. The dispute centred on a later sale. In 2016, the Andy Warhol Foundation (AWF) licensed one of those works, "Orange Prince," to Condé Nast to illustrate a magazine story about Prince. AWF received $10,000. Goldsmith received nothing. After Goldsmith asserted her copyright, AWF sued for a declaratory judgment that the Prince Series did not infringe because it qualified for fair use, and Goldsmith countersued for infringement.

The district court sided with the foundation. On July 1, 2019, the court granted summary judgment for the Warhol Foundation, finding that the Prince Series was fair use. Its reasoning rested on the idea that Warhol recast Prince as an "iconic, larger-than-life figure," while the Goldsmith photograph portrayed him as a "vulnerable human being." The Second Circuit reversed in 2021. A three-judge panel unanimously held that the Prince prints did not qualify as transformative works of art. It also held that the artist's intent, and a critic's or judge's reading of meaning, could not by themselves establish transformation. The court found that all four fair use factors favored Goldsmith. The Supreme Court later took up only the first factor, and because AWF did not challenge the Second Circuit's holding on factors two through four, the only issue before the Court was the first factor.

On May 18, 2023, the Court ruled 7–2 for Goldsmith, with an opinion by Sonia Sotomayor. The central holding is that the mere addition of "some new expression, meaning, or message" is not enough to tip the factor toward fair use. The Court framed the inquiry as a question of degree. It asks whether the challenged use has a further purpose or a different character from the original, and how far that difference goes, rather than whether the secondary work looks different or says something new. The majority stressed that its ruling was limited to the narrow issue of the first factor, and to the challenged use. On the facts, the Court found that the two images shared the same purpose: both were featured in magazines as portraits of the same musician. Because AWF's use was commercial, the factor weighed against it. The Court also reasoned that loosely defined transformation would hollow out the exclusive right to prepare derivative works, since ordinary adaptations such as arrangements, sequels, and stage or film versions could be treated as transformative. If transformation alone were sufficient, that would endanger the copyright holder's exclusive right to create derivative works.

After the decision, commercial licensing and substitution play a much larger role in both factor one and factor four. If an original work and a secondary use share the same or highly similar purposes, and the secondary use is of a commercial nature, the first factor is likely to weigh against fair use, absent some other justification for copying. The Second Circuit had already linked the analysis to the market: the district court correctly found that the primary markets for the two works did not meaningfully overlap, but the appeals court concluded that the Prince Series could harm the photographer's market for licensing her image. Commentators have therefore described the decision as one that makes it more likely that the first factor will fail when both works are similar and have a significant commercial purpose.

Scholars and practitioners disagree about how far the ruling reaches. Some view it as a narrow decision about magazine licensing of a portrait. One art lawyer quoted in commentary called it "narrow", and the majority "took pains to note that other works by Warhol had little to fear from the Copyright Act." Others argue that the narrowness is overstated. Nothing in the opinion states that the broad analysis of the first fair use factor is itself limited. Critics of the majority also note that the decision leaves the art field with inadequate guidance, because the Court did not squarely analyze larger questions about how fair use should apply to appropriation art generally. Justice Kagan's dissent, joined by Chief Justice Roberts, argued that the majority gave too little weight to what Warhol added and that the ruling would discourage creative work. The debate therefore turns on whether the decision mainly affects commercial appropriation, or whether it changes how transformative use is weighed across visual art, music, literature, and other fields.

Fair dealing and other national models

Fair dealing in the United Kingdom, Canada, and Australia

Fair dealing grew out of a judge-made doctrine in the British courts of the nineteenth century. The judge-made fair use doctrine was codified in the United Kingdom in 1911 and in Canada ten years later. The statutory fair dealing provisions in the UK and Canada set out a closed list of permitted purposes. In the UK, the categories are noncommercial research and private study, criticism and review, as well as the reporting of current events or official proceedings, and the user must give sufficient acknowledgment of the author and title of the quoted work. Australia's provisions are generally similar with respect to research or study, criticism or review, or news reporting, with later additions such as parody and satire.

Canada has gone further than the older British list. The Copyright Modernization Act, 2012 added the fair dealing purposes of education, parody and satire to a list that already included research, private study, criticism, review, and news reporting. Canadian law also imposes a source-attribution condition: if you copy something for the purpose of criticism, review, or news reporting, you must acknowledge your source; otherwise the dealing will categorically not be fair.

Expansive readings in Canada

The Supreme Court of Canada's 2004 decision in CCH Canadian Ltd. v. Law Society of Upper Canada changed how the closed list operates. The Court characterized the fair dealing exceptions as a user right, which must be balanced against the rights of copyright owners. It also held that "research" must be given a large and liberal interpretation in order to ensure that users' rights are not unduly constrained. Later decisions applied the same generous approach to every enumerated purpose, not only to research. In practice, a liberal interpretation of fair dealing led in the iTunes Case to democratizing "research" to include not just what academics and scientists do but also research carried on by ordinary members of the public, and in the School Photocopying case, teachers' copying for students was treated as falling within the students' purpose of study. Commentators have concluded that virtually all substantive dealings with a work will qualify for one of the categories of fair dealing, which brings Canada's fair dealing rights much closer to American fair use rights. Critics, however, have described the closed list as illogical and too constraining, and the broader reading has also created uncertainty about its scope.

Civil-law systems and the three-step test

Civil-law countries generally take a different structural approach. Their copyright statutes contain specific, enumerated exceptions rather than a general user right. The copyright laws in many "civil law" countries currently allow their courts to apply a specific exception in a specific case only if second and third steps of the Berne three-step-test are met. Under those steps, the court may permit the use only if it determines that the use does not conflict with a normal exploitation of the work and does not unreasonably prejudice the legitimate interests of the rights holder. The commentary notes that these steps are at least as abstract and difficult to apply as fair use or fair dealing, so the model does not fully escape the problem of open-ended judgment.

The Berne Convention itself lists specific permitted uses. Among them are private study, quotation, and teaching and research uses, which appear in its articles 9(2), 10(1), and 10(2). Fair practice under Article 10 is a distinct concept from fair use or fair dealing. Fair practice is a narrower standard tied to a specific use, such as quotation, and national laws often reproduce it in those provisions. Botswana's 2000 Act, for example, refers to fair practice in its quotation provision and in its provision on reproduction for teaching, and the latter is limited to a 'short part of a published work'.

Replacement of fair dealing by specific exceptions

Some former British colonies have moved away from fair dealing entirely. Four former colonies, Botswana, Ghana, Lesotho, and Malawi, have replaced fair dealing with other exceptions. Botswana's experience shows how this happened. When Botswana gained its independence in 1966, the 1956 Act remained in force for over three decades. Following Botswana's accession to international copyright agreements such as the World Trade Organization (WTO) Agreement on Trade-Related Aspects of Intellectual Property Rights (TRIPS) in 1998, a new Copyright and Neighbouring Rights Act was adopted in 2000. That Act contains a range of specified uses (nine in total) including personal use, teaching and libraries and archives. Ghana followed a similar path: when Ghana adopted its new Copyright Act in 2005, it replaced fair dealing with other exceptions.

The trade-off is between predictability and flexibility. A closed set of predetermined uses gives users clearer rules, but it can be hard to update. Specific exceptions can enable the law to adapt more quickly to technological change or to accommodate uses not foreseen when the law was developed. Yet even precise statutory wording can produce doubt. A study of Botswana's library exception concluded that the clause in question is uncertain, unclear and ambiguous, and needs to be revised.

International norms and harmonization

The Berne three-step test

The international benchmark for copyright exceptions is the three-step test in Article 9(2) of the Berne Convention. It is often called the three-step test, and commentators treat it as the benchmark for judging exceptions to exclusive rights. Article 9(2) leaves it to national legislatures to allow copying in limited circumstances, but only where the copying does not conflict with a normal exploitation of the work and does not unreasonably prejudice the legitimate interests of the author. The provision was not part of the original Berne text. It was inserted only at the time of the Stockholm Revision in 1967, and it appeared in the 1971 Paris Act.

The formula is vague, so scholars have split it into separate tests. One formulation reads the three steps as bars on exceptions that are overly broad, on exceptions that strip rightsholders of a substantial source of income, and on exceptions that cause disproportionate harm. The three-step test in Article 9(2) of the Berne does not apply to copyright exceptions that are implemented under other parts of the Berne convention, so Berne's other exceptions follow different rules. The test also has a narrow reach in Berne itself: it applies only to the right of reproduction and not to public performance, broadcasting, adaptation, or the rights granted to producers of cinematographic works.

Quotation and illustration under "fair practice"

Berne also contains exceptions that are judged by a different standard. Quotation and use for teaching are permitted when they are compatible with "fair practice," a phrase that the Convention uses instead of the three-step language. Commentators read these provisions as operating independently of Article 9(2), so national quotation or teaching rights need not separately pass that test. Many national laws have converted this standard into explicit purposes. Israel's 2007 statute, for example, lists quotation and instruction among the purposes it covers, and Israel's fair use provision, Section 19 of the Copyright Law 2007, is a close translation of Section 107 of the U.S. Copyright Act.

Trade law and the WTO

The three-step formula moved into trade law through the TRIPS Agreement. It also extended in its Article 13 the three-step test of the Berne Convention to cover any copyright right (including, for example, public performance). The WTO dispute settlement system first applied the test in 2000. In the dispute, initiated by the European Union at the behest of the Irish performing rights organization, the contested exception, enacted in the 1998 "Digital Millennium Copyright Act," exempted a broad range of retail and restaurant establishments from liability for the public performance of musical works. The panel found that the United States had held the United States in contravention of its obligation under art. 13 of the TRIPs accord.

The panel's approach was restrictive. This interpretation results in a restrictive reading of the three-step test, requiring parties to independently satisfy each of the three criteria. Its treatment of the first step also stressed narrowness: the first condition of Article 13 requires that a limitation or exception in national legislation should be clearly defined and should be narrow in its scope and reach. The formula also differs from its trademark and patent counterparts, because in the WTO versions of the trademark and patent 3-step tests, the interests of third parties are given explicit status to balance the interests of right holders. The copyright 3-step test does not mention third parties.

The World Intellectual Property Organization has carried the test into its own treaties. The three-step test is embodied not only in the Berne Convention but also in the TRIPs Agreement and in the WIPO internet treaties, including article 10 of the WIPO Copyright Treaty. WIPO has also hosted long-running committee discussions on exceptions for libraries, archives, and educational institutions, but no binding WIPO instrument on those exceptions has been adopted.

Countries that adopted US-style fair use

Fair use remains rare outside the United States. US-style fair use is used in a small but growing group of countries—the Philippines, Israel, South Korea and Singapore among them. Counts vary with definitions. One 2018 article lists Israel, Liberia, Malaysia, the Philippines, Singapore, South Korea, Sri Lanka and Taiwan [as having adopted] the fair use regime or its close variants. The models also differ in drafting. The Philippines' Intellectual Property Code provision is virtually identical to Section 107 of the U.S. Copyright Act, while Singapore's version is a "fair dealing" exception that is structured as an open-ended fair use exception. By contrast, the fair use law of countries such as Israel sets out a list of uses that is understood to impose some outside limit on which types of uses may qualify as fair use.

Adoption remains contested. Other nations, including Australia, New Zealand, South Africa, Nigeria and Hong Kong, are debating whether to adopt fair use or something like it. Opponents in Australia argue that the reform would add uncertainty without need. Supporters of transplantation cite economic evidence: a preliminary dataset submitted in a US trade docket associates the adoption of fair use clauses modeled on U.S. law with positive outcomes for the firms in the dataset. That evidence is limited and disputed.

Limits of harmonization

Harmonizing fair use is difficult for structural reasons. All copyright law is national law; it applies to people and companies within that country. Because the specificity of copyright exceptions based upon and consistent with the three-step test varies significantly, for example between the US and its unspecific Fair Use exemption and the European Unions exhaustive and comparably specific list of potential limitations, scholars have questioned whether an open-ended standard can meet a test designed for narrow exceptions. Whether US fair use is compatible with Berne and TRIPS obligations has been questioned many times over the years by European and even American commentators.

The question is now prominent in AI policy. Commentators argue that any (new) exception for the use of copyrighted content as AI input, whether on an international or national level, should therefore comply with the three-step test. Whether open-ended fair use or closed, enumerated exceptions can meet that standard remains an open question across jurisdictions.

Education, libraries, and research

Classroom and university use

Teaching and scholarship appear among the purposes the US statute names as potentially fair, and universities rely on the doctrine heavily. Institutions commonly use it to support excerpts in course readers, chapters assigned for seminars, and images, audio, and video embedded in lectures or online course sites. These uses are usually non-commercial and tied to instruction, which tends to help the user under factor one. Disputes often concentrate on factor four, because a course pack that replaces purchases of a textbook affects a market that publishers can exploit through licensing.

Multimedia materials raise separate questions. Showing a film clip in a seminar room is usually assessed differently from distributing a copy of that clip to students or the public. Film and media educators have drawn up their own guidance on this point: a code of best practices for film and media educators, designed by the Society for Cinema and Media Studies, deals with classroom screenings, broadcasts, and derivative works.

Libraries and archives

In the United States, libraries also operate under narrow statutory privileges in section 108 of the copyright law, which sit alongside fair use. For interlibrary loan, a library may make one copy of an item it holds in order to lend it to another library. For preservation, up to three copies of an unpublished work held by the library may be made for preservation purposes. Digital preservation copies are restricted: if the copy is digital, it cannot be circulated outside the library.

These provisions are deliberately limited. The receiving institution may not retain any copy made under these provisions in order to augment its collections or to facilitate further ILL. The Copyright Office has also argued that the exceptions did not anticipate and no longer address the ways in which copyrighted works are created, distributed, preserved, and accessed in the twenty-first century. Comparable library and archive exceptions exist in many national laws, though their scope differs considerably.

Text and data mining

Text and data mining (TDM) raises a distinctive question: the analytical goal is to derive patterns rather than to reproduce the works themselves, yet a computer must first copy the material. In the European Union, TDM is defined as an automated analytical technique aimed at analysing text and data in digital form in order to generate information such as patterns, trends and correlations.

The EU's 2019 copyright directive separates two exceptions. Article 3 covers research organisations and cultural heritage institutions, which may extract and reproduce text and data from databases or other sources they lawfully access, for scientific research. Commercial applications fall outside it. Article 4 is broader in its beneficiaries: it permits reproductions of and extractions from lawfully accessible works for TDM for any purpose, provided rightsholders have not expressly reserved that activity in an appropriate manner. Copies made for mining are expected to be kept under secure conditions, as copies generated by text and data mining should be stored securely.

The distinction between analysing and reproducing matters in practice. Courts and legislators tend to treat the copying step as the act that needs authorization, even when the output contains only facts or statistical features. This is why TDM provisions usually focus on the making of copies, the retention of copies, and the rightsholder's ability to opt out, rather than on the research result.

Codes of best practice

Beyond statutes and case law, several communities have drafted codes of best practice that translate fair use into shared working rules. Documentary filmmakers released their statement of best practices in fair use on November 18, 2005. The Association of Research Libraries published its code for academic and research librarians, not all librarians, in January 2012. Among the eight other communities of practice that established codes of best practices in fair use for themselves between 2005 and 2012, several reported that the exercise helped them use their rights with greater confidence.

These codes carry no legal force of their own. They matter because courts have long considered whether the user acted reasonably and in good faith in light of standards of accepted practice in his or her particular field. A documented community standard therefore gives educators, librarians, and researchers evidence of what reasonable practice in their field looks like, which is one way of reducing the uncertainty that the doctrine otherwise leaves to case-by-case judgment, a concern that knowledge-production institutions have raised repeatedly.

Digital technology, search, and AI training

Home recording and consumer technology

The Supreme Court's 1984 decision in Sony Corp. of America v. Universal City Studios (the Betamax case) held that households recording broadcast television to watch later, known as time-shifting, were making fair use. The Court also held that selling a recording device with substantial non-infringing uses did not make its maker liable for users' infringement. That reasoning gave later courts a template for consumer technology: if a product has lawful uses, the analysis shifts from the device to the conduct of particular users. Later disputes over software, storage, and online services often drew on this framework.

Search, caching, and indexing

Early web cases applied factor one to copies made to locate and organize information rather than to provide the original work's expressive content. In the Ninth Circuit's Kelly v. Arriba Soft (2003) and Perfect 10 v. Google (2007) decisions, reduced-size thumbnails in image search results were treated as transformative because they served an informational function. A Nevada district court in Field v. Google (2006) found that cached copies of web pages were fair use, partly because the plaintiff had not used the available means to block crawling. The same logic explains why indexing, which stores data about works, is generally treated more leniently than displaying full copies.

Training generative AI

Copyright disputes over AI now separate several acts: acquiring works, retaining copies, training, retrieving material, distributing models, generating outputs, and user conduct. Courts must separate acquisition, retention, training, retrieval, model distribution, outputs, and user conduct. Each act can face a different fair-use analysis. A fair-use ruling for training does not immunize model outputs. The central questions are whether ingesting works to learn statistical patterns is transformative, whether models that memorize and reproduce passages infringe, and whether outputs that compete with the originals harm the market under factor four.

District courts have reached different results. Judge Alsup's 2025 order in Bartz v. Anthropic described training Claude on books as fair, calling the technology "spectacularly" transformative. The same court ruled that AI training on copyrighted books constitutes fair use, but storing pirated copies does not. That case later settled for US$1.5 billion, with an estimated payout of approximately US$3,000 per work. In Kadrey v. Meta, the court ruled for Meta on the record presented, but warned that plaintiffs with stronger market-dilution evidence might fare better in future cases.

Appellate rulings and open litigation

On 29 September 2026, the Third Circuit affirmed Thomson Reuters' partial summary judgment against ROSS Intelligence in the first federal appellate decision to address fair use in AI training. The court found that ROSS's copying was not fair use, and the analysis turned mainly on the first and fourth factors. The ruling is narrow in scope. Unlike the AI models in Bartz and In re OpenAI, ROSS's AI platform cannot generate original expression, and therefore its use is not transformative. Commentators stress that the court repeatedly emphasized the case should not be viewed as a broad ruling on generative AI training.

Much remains unresolved. The Second and Ninth Circuits have not spoken on fair use in AI training, and the possibility of a circuit split on the scope of fair use for AI training remains open, and the Supreme Court has not yet spoken to the question directly. In the New York Times v. OpenAI consolidated litigation, cross-motions for summary judgment were filed in the OpenAI MDL on September 4, 2026, but no fair-use ruling has yet been issued.

Policy and licensing proposals

Legislators and regulators are pursuing other tools. In the United States, House Representatives Madeleine Dean and Nathaniel Moran introduced H.R. 7209 on January 22, 2026, which would let copyright owners seek subpoenas to learn whether their works were used to train AI models. In the United Kingdom, the government said it will not introduce any reforms to copyright law after consultation, having earlier proposed a text and data mining exception with an opt-out. The European Union has taken a different route: Article 53 of Regulation (EU) 2024/1689—the EU AI Act—now imposes enforceable copyright obligations on all GPAI providers. A non-binding European Parliament resolution from March 2026 has also proposed a flat-rate licensing fee of 5 to 7 percent of global turnover for creative industries. Whether AI training should rely on open-ended fair use, statutory exceptions, or licensing remains contested.

Criticism, uncertainty, and open questions

The most frequent objection to fair use is that its outcomes are hard to predict. Because the doctrine turns on a fact-specific weighing of four factors, legal-precedent offers only partial guidance, and a use that one court treats as fair may be treated differently by another court or in a different record. Scholars who defend the doctrine answer that the case law is more coherent than critics assume, and that common patterns emerge when decisions are grouped by the kind of use involved. Even so, the practical effect is that creators, publishers, platforms, and users often cannot know whether a use is lawful until a dispute has been litigated. Critics argue that this uncertainty chills expression that would be legitimate under a clearer rule, while supporters argue that the same flexibility is what allows copyright to adapt to new technologies and forms of commentary.

Costs of litigation and statutory damages

In the united-states, the risk of uncertainty is magnified by the remedies available for infringement. Copyright owners may elect statutory damages instead of proving actual losses, and the amount per work can reach very high levels when infringement is found to be willful. Critics note that a fair use defense that may succeed only after lengthy and expensive proceedings is a weak protection for small creators, independent journalists, and individual users, who may settle rather than fight. Some commentators contend that the combination of a case-by-case defense and large, unpredictable awards pushes intermediaries to over-remove material, a concern raised by civil-liberties organizations. Proposed responses include limiting statutory damages where a defendant has a good-faith fair use claim, or creating clearer safe harbors for particular categories of use. Supporters of the current system respond that these proposals would weaken protection for owners whose works are copied at scale.

Transformative purpose and market harm

A second line of criticism concerns the weight given to transformation under the first factor. Some scholars argue that courts have sometimes treated a new purpose or added meaning as decisive, allowing that finding to overshadow the fourth factor, which asks whether the use harms the market for the original or its licensed derivatives. On this view, a use can look transformative in the abstract while still substituting for the work in the market where the owner earns revenue. The 2023 Supreme Court decision in the Warhol dispute is often cited as a corrective to that tendency, although its reach is debated. Defenders of broad fair use reply that open-ended standards are necessary for innovation, criticism, and new forms of expression, and that a narrower reading of transformation would leave many socially valuable uses without protection. The dispute therefore reflects a deeper disagreement about whether fair use should be read as a mechanism for protecting expression or as a limit on the private control of information.

Recent appellate litigation over artificial-intelligence illustrates the stakes. In September 2026 the Third Circuit affirmed a ruling that a startup's copying of legal headnotes to train a competing legal research tool was not fair use, and it described that decision as the first federal appellate ruling on fair use in AI training. Commentators stress, however, that the court addressed a non-generative system that competed directly with the plaintiff, and that its reasoning does not settle whether training a highly transformative generative model is lawful. In earlier 2025 district court decisions, judges reached different conclusions on related facts, and one emphasized that its outcome depended on the particular evidentiary record.

Unresolved questions

Several issues remain open. Cross-border enforcement is difficult because fair use is a US doctrine, while most other jurisdictions rely on civil-law exceptions, fair dealing, or statutory lists, so a use lawful in one country may infringe in another. The treatment of unpublished works remains sensitive, since the second factor gives them heightened protection even where the use is otherwise defensible. Orphan works, whose owners cannot be located, pose a related problem: a user may be willing to pay a license fee yet have no one to pay, and the fourth factor offers little guidance when no market can be identified. Finally, debate continues over whether open-ended fair use or specific statutory exceptions better serve the public interest. Fair use offers flexibility for unforeseen uses but limited predictability; specific exceptions offer certainty but may lag behind technology and may need to be drafted narrowly enough to exclude many legitimate uses. How these trade-offs are resolved will likely depend on future appellate decisions, legislative proposals on AI training and licensing, and the development of professional practices for everyday use.

References

1. Fair Use - Copyright and Fair Use Guide - Research Guides at Sam Houston State Universityshsulibraryguides.org
2. 17.23 Copyright—Affirmative Defense—Fair Use (17 U.S.C. § 107)www3.ce9.uscourts.gov
3. Fair Use Exception To Copyrightcopyrightalliance.org
4. What Is Fair Usecopyrightalliance.org
5. Fair Use, the Four Factors, and When It Doesn't Protect Youfasthofflawfirm.com
6. 17 U.S.C. § 107: Four Factors, Burden of Proof, and AI Training - FedLawsfedlaws.org
7. Copyright and Fair Useogc.harvard.edu
8. Fair Use: The 4 Factors Courts Consider in a Copyright Infringement Casenolo.com
9. (PDF) The Pre-History of Fair Useresearchgate.net
10. Folsom v. Marsh - Citizendiumen.citizendium.org
11. Commentary on: Folsom v. Marsh (1841)copyrighthistory.org
12. Fair Use Week (Sec. 107)authorsalliance.org
13. GENERAL GUIDE TO THE COPYRIGHT ACT OF 1976 SEPTEMBER 1977copyright.gov
14. Fair useen.wikipedia.org
15. Appendix Two: Fair Use Then and Nowopenpress.wheatoncollege.edu
16. Harper and Row v. Nation Enterprises (1985)firstamendment.mtsu.edu
17. Fair Use - The Fordham Law Archive of Scholarship and Historyir.lawnet.fordham.edu
18. Folsom v. Marshen.wikipedia.org
19. fair use historymatthewsag.com
20. Circuit Court, D. Massachusetts. Oct. Term, 1841. FOLSOM ET AL. V. MARSH ET AL.law.resource.org
21. Harper & Row, Publishers, Inc. v. Nation Enterprises, 471 US 539, Supreme Court 1985, (BitLaw)bitlaw.com
22. Harper & Row v. Nation Enterprisesen.wikipedia.org
23. Harper & Row Publishers, Inc. v. Nation Enterprises, 471 U.S. 539 (1985): Case Brief Summaryquimbee.com
24. 17 U.S. Code § 107 - Limitations on exclusive rights: Fair uselaw.cornell.edu
25. Harper & Row v. Nation Enterprisesitlaw.fandom.com
26. Andy Warhol Foundation Visual Arts v. Goldsmith, 598 US 508, Supreme Court 2023, (BitLaw)bitlaw.com
27. Harper & Row v. Nation Enterprises, 471 U.S. 539 (1985).tile.loc.gov
28. Harper & Row, Publishers, Inc. v. Nation Enterprisesstudicata.com
29. Harper & Row, Publishers, Inc. v. Nation Enterprisescourtlistener.com
30. Harper & Row v. Nation Enterprises - Sandra Day O'Connor Institute Librarylibrary.oconnorinstitute.org
31. ANDY WARHOL FOUNDATION FOR THE VISUAL ARTS INC v. GOLDSMITH (2023)caselaw.findlaw.com
32. Foundation Models and Fair Usearxiv.org
33. ANDY WARHOL FOUNDATION FOR VISUAL ARTS, INC. v. GOLDSMITHlaw.cornell.edu
34. Campbell v. Acuff-Rose Music, Inc. (1994)firstamendment.mtsu.edu
35. Parody and Fair Use: Campbell v Acuff-Rose Music (1994)h2o.law.harvard.edu
36. Campbell v. Acuff-Rose Music, Inc.en.wikipedia.org
37. API Software Licensing After Google v. Oraclebarclaydamon.com
38. Google v. Oracle: Supreme Court Rules for Google in Landmark Software Copyright Casecongress.gov
39. The Supreme Court’s Decision in Google v. Oracleclearygottlieb.com
40. Second Circuit Affirms Fair Use in Google Books Case — Association of Research Librariesarl.org
41. The Authors Guild v. Google, Inc.loeb.com
42. Authors Guild v. Google, Inc., 804 F. 3d 202 - Court of Appeals, 2nd Circuit 2015 - Google Scholarscholar.google.com
43. Campbell v. Acuff-Roseblogs.law.gwu.edu
44. SCOTUS Says 'Fair Use' in Google v. Oracle Copyright Battlemcdermottlaw.com
45. Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmithoyez.org
46. 2nd Circuit Flays Fair-Use Defense to Warhol Prince Printscourthousenews.com
47. Andy Warhol Foundation for Visual Arts, Inc. v. Goldsmithsupreme.justia.com
48. TOP STORY—2d Cir.: Warhol’s ‘Prince...vitallaw.com
49. A Blow to Pop Art: Case Review of Warhol v. Goldsmith (2021) - Center for Art Lawitsartlaw.org
50. Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmithlaw.cornell.edu
51. U.S. Supreme Court Warhol Foundation Decision on Copyright Fair Usemsba.org
52. Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith (21-869)scotusblog.com
53. Supreme Court Clarifies Copyright Fair Use Defense in Andy Warhol Foundation v. Goldsmithgoodwinlaw.com
54. Dissent Says SCOTUS Ruling Against Warhol Foundation on Fair Use Will 'Stifle Creativity'ipwatchdog.com
55. Everything might be OK! Warhol v. Goldsmith - Creative Commonscreativecommons.org
56. Andy Warhol, Commercialism, and The Doctrine of Fair Usecrowell.com
57. Grossman LLPgrossmanllp.com
58. Landmark Warhol Decision Reins in Transformative Fair Usecopyrightalliance.org
59. Decision Alert: Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmith, No. 21-869mayerbrown.com
60. Client Alert: Putting the “Use” Back in Fair Use: The Supreme Court Decides Andy Warhol Foundation for the Visual Arts, Inc. v. Goldsmithjenner.com
61. Andy Warhol's 'Prince Series' of Silkscreens Is Not Fair Use, Federal Appeals Court Ruleslawandcrime.com
62. The Supreme Court Strikes a New Chord in Andy Warhol Foundation v. Goldsmithaxley.com
63. From Art Museums to the Supreme Court: How Does the Decision in Warhol v. Goldsmith Go Beyond Art?libraryfutures.net
64. Did the Supreme Court Transform Its Transformative Use Test In Warhol?ficlaw.com
65. Appendix One: Fair Dealing in Canadapressbooks.library.torontomu.ca
66. Legal Definitionsjoycefoundation.utulsa.edu
67. Fair dealing in Canadian copyright lawen.wikipedia.org
68. User Rights Then and Nowcaut.ca
69. CCH Canadian Ltd v Law Society of Upper Canadaen.wikipedia.org
70. Celebrating 20 Years of Fair Dealing: Reflecting on the Landmark Decision of CCHcippic.ca
71. Healing Fair Dealing? A Comparative Copyright Analysis of Canada's Fair Dealing to U.K. Fair Dealing and U.S. Fair Usecanlii.org
72. The Fair Use/Fair Dealing Handbookinfojustice.org
73. EIFL’s Review of ‘Fairness’ in Partner Countries’ Copyright Lawsinfojustice.org
74. Fair dealing: to replace or reinstate?eifl.net
75. (PDF) Limitations & Exceptions under Copyright Law in Relation to Libraries & Archives Botswana Copyright & Neighboring Rights Act 2000researchgate.net
76. Fair dealingen.wikipedia.org
77. The Fair Use/Fair Dealing Handbook Jonathan Band Jonathan Gerafiinfojustice.org
78. Limitations and Exceptions in Copyright Law for Educational andwipo.int
79. The three-step test, deemed quantities, libraries and closed exceptionsstatic-copyright-com-au.s3.amazonaws.com
80. Berne three-step testen.wikipedia.org
81. An Overview of the International Treatment of Exceptionsdigitalcommons.wcl.american.edu
82. Transplanting Fair Use Across the Globe: A Case Study ...core.ac.uk
83. The Reverse Three-Step Test - Scholarship@Vanderbilt Lawscholarship.law.vanderbilt.edu
84. Toward Supranational Copyright Law? The WTO Panel Decision and the 'Three-Step Test' for Copyright Exceptionsresearchgate.net
85. United States Four Fair Use Factors and the WTO Three-Step Testkeionline.org
86. The three-step test within the copyright systempcf4.dec.uwi.edu
87. Three-Step test in copyright negotiationskeionline.org
88. A Scientific Analysis of the Three-Step Test: Through the Lenses of International and Australian Lawslink.springer.com
89. Negotiating Copyright Exceptions across Bordersdocumentary.org
90. Document Typescholarship.law.tamu.edu
91. Firm Performance in Countries With and Without Open Copyright Exceptionsdownloads.regulations.gov
92. Is the U.S. Fair Use Doctrine Compatible with Berne and TRIPS Obligations? by Pamela Samuelson, Kathryn Hashimoto :: SSRNpapers.ssrn.com
93. Who Owns the Output? Bridging Law and Technology in LLMs Attributionarxiv.org
94. THE FAIR USE/FAIR DEALING HANDBOOKdigitalcommons.wcl.american.edu
95. History of fair use proposals in Australiaen.wikipedia.org
96. Exceptions to Intellectual Property Rights: Lessons from WTO-Trips Panels - [2003] MurUEJL 31www5.austlii.edu.au
97. Toward Supranational Copyright Law? The WTO Panel ...scholarship.law.columbia.edu
98. Codes of Best Practice - Copyright - LibGuides at Tennessee State Universitytnstate.libguides.com
99. Exceptions for Libraries and Archives - Copyright and Fair Use - LibGuides at Oregon State Universityguides.library.oregonstate.edu
100. Section 108 of Title 17copyright.gov
101. Revising Section 108: Copyright Exceptions for Libraries and Archivescopyright.gov
102. Text and Data Mining: (Articles 3 and 4 of the EU-DSM) by REBIUN’s Copyright working groupblogs.ifla.org
103. Mining for exceptions in the new copyright directive - Lexologylexology.com
104. Copyright: Text and Data Miningkempitlaw.com
105. 1 How American Filmmakers Created a Code of Best Practices in Fair Usecmsimpact.org
106. CODE OF BEST PRACTICES IN FAIR USE FOR ACADEMIC AND RESEARCH LIBRARIESarl.org
107. AI Copyright Litigation Is The Biggest Active Fightlicentium.io
108. The Open Questions in U.S. Generative AI Copyright Litigationclearyiptechinsights.com
109. AI in litigation series: An update on AI copyright cases in 2026nortonrosefulbright.com
110. AI Training Fair Use Loses First Appellate Test; Generative AI Gets Explicit Carve-Outtechtimes.com
111. Third Circuit Addresses Fair Use in AI Training, But Leaves Generative AI Questions Unresolvedballardspahr.com
112. First Federal Appellate Fair Use Ruling on AI Training Rejects Non-Generative AI’s Use of Copyrighted Headnoteswomblebonddickinson.com
113. Third Circuit Affirms Copyright Protection for Westlaw Headnotes and Rejects Fair Use Defense in AI Training Caseharrisbeachmurtha.com
114. Fair use and AI training: what courts have actually ruledailawsuittracker.com
115. New House Bill on AI Transparency Aims to Pull Back the Curtain on AI Training Datanatlawreview.com
116. Government drops plans to introduce copyright exception for AI training - Museums Associationmuseumsassociation.org
117. EU Copyright & Generative AI Training Data Report 2026aigovernancedesk.com
118. White Paper on Remixes, First Sale, and Statutory Damagesuspto.gov
119. 2537 UNBUNDLING FAIR USES Pamela Samuelson*fordhamlawreview.org
120. Market Design for AI: Beyond the Copyright Binaryarxiv.org
121. Statutory Damages: The Fuel of Copyright-based Censorshipeff.org
122. Thomson Reuters v. ROSS: Appeals Court Rejects AI Fair Useen.ai-creators.tech
123. First Federal Appellate Fair Use Ruling on AI Training Rejects Non-Generative AI’s Use of Copyrighted Headnotesnatlawreview.com
124. AI training of copyrighted material not fair use: Third Circuitcourthousenews.com
125. The Illegality of AI Training and the Limits of Fair Use in Copyright (Thomson Reuters v ROSS Intelligence)technologylaw.ai