AI Is Reigniting Decades-Old Questions Over Digital Rights, but Fair Use Prevails

Katherine Klosek, director of Information Policy and Federal Relations at the Association of Research Libraries, and Samantha Teremi, the Licensing Librarian at UC Berkeley Library, examine how today’s contractual restrictions on AI echo the same concerns libraries raised 20 years ago during the US Copyright Office Digital Millennium Copyright Act (DMCA) Section 104 study. This piece is cross-posted on the Harvard University Fair Use Week blog.

More than two decades ago, libraries in the United States raised concerns about licenses for digital works restricting fair use and other library rights as part of the US Copyright Office DMCA Section 104 study. ARL and other library associations explained how restrictive licensing terms and pay-per-use models force institutions to accept limiting conditions to provide essential resources for faculty and students or pay to perform certain functions with content that was already licensed and paid for. The libraries made the case that fair use rights are as important in the new digital environment as they are in the print world and that federal copyright law should preempt restrictive license terms. At the time, the Copyright Office suggested that there was no current evidence of the problem of license terms determining the landscape of library and user rights. Unfortunately, those early warnings of how restrictive licensing terms and pay-per-use models can impede research proved prescient.

Given the ongoing legal battles over fair use and copyright in the present, it is time to update our collective understanding of how licenses for digital content benefit those who set the terms and increasingly erode the fair use rights of libraries and other users of copyrighted works.

New Technology, Same Problems

The fundamental problem of licenses undermining libraries and scholarly research remains, even as technology and business models shift. Today, some publishers are attempting to implement new contractual bans on artificial intelligence by sending impromptu addendums to their customers and, in some cases, claiming that they’re unable to waive these bans because they intend to create their own AI tool and ultimately license AI usage rights back to their customers for an additional cost.

For instance, a publisher recently provided UC Berkeley’s Library with an elusive explanation for its AI ban on a subset of its licensed materials, claiming that it would require new and different AI terms that would be significantly higher in price and that individual client requests would need to be evaluated to determine whether or not they would be permitted. However, when prompted to provide said new terms and price, the publisher was unable, or perhaps unwilling, to provide any additional information, noting that there was no set pricing model or terms to share. This publisher also stated that up to that point, it had not received a client request for that level of AI use. Yet, in speaking with colleagues about this publisher, librarians at UC Berkeley learned that other libraries had also inquired about AI rights, but these tactics discouraged them from continuing their pursuit, especially when faced with a publisher who was not particularly willing to come to the table nor forthcoming about their company’s AI concerns.

Charging extra to secure AI rights is likely to be cost-prohibitive due to increased financial burdens on libraries and institutions of higher education; if publishers are successful, it could lead to less academic output as researchers may have to independently foot the bill for the right to conduct research using AI. To support libraries in forming a united front to preserve fair use rights, UC Berkeley Library shares language and strategies to help libraries successfully negotiate contracts supporting research rights through clauses that address vendors’ concerns around information security and market share without restricting scholarly use.

Non-negotiated Contracts Interfere with the Right of Fair Use

Some publishers attempt to restrict the use of AI on their content through non-negotiated contracts like browse-wrap licenses, end-user license agreements (EULAs), or terms of service, which are legally binding once users click a button to accept a website’s terms and conditions. The New York Times updated its terms of service to prohibit scraping articles and images for AI training, and other websites are sure to follow. Even though consumers rarely read these terms, courts treat them as negotiated contracts between sophisticated parties.

In a promising legal development, a federal court in California recently ruled that non-negotiated contracts may not be able to restrict otherwise lawful access to copyrighted works. In the 2024 case X Corp. v. Bright Data, the US District Court for the Northern District of California found that contractual prohibitions on scraping web content imposed by X (formerly Twitter) interfered with the exercise of the statutory privilege of fair use. We hope other courts will follow the conflict-preemption analysis in cases when non-negotiated contracts interfere with fair use and other rights.

While a federal legislative solution to the problem of non-negotiated contracts interfering with fair use is unlikely in the near future, there is a model in the 2002 Digital Choice and Freedom Act. The bill, introduced by Representative Zoe Lofgren (D-CA), would have created a new section of the US Copyright Act asserting that license terms that restrict any of the limitations on the copyright holder’s exclusive rights are not enforceable under any state statute.

(b) EFFECT OF LICENSES.—When a digital work is distributed to the public subject to nonnegotiable license terms, such terms shall not be enforceable under the common laws or statutes of any State to the extent that they restrict or limit any of the limitations on exclusive rights under this title.

This clause would protect the rights of libraries and other consumers of digital works by voiding nonnegotiable license terms that restrict fair use and other limitations and exceptions in US copyright law.

Copyright Warnings Do Not Have the Force of Law

If you’ve watched a broadcast of a professional baseball game, you’ve likely heard the disclaimer, “Any rebroadcast, retransmission, or account of this game, without the express written consent of Major League Baseball, is prohibited.” Like FBI warnings on VHS tapes and DVDs, these copyright warnings do not have the force of law. Rightsholders use them to intimate users into asking permission—and usually paying—to use their content, without regard for fair use.

In 2007, the Library Copyright Alliance (LCA) supported a complaint filed by the Computer and Communications Industry Association (CCIA) with the US Federal Trade Commission (FTC) contending that overbroad copyright warnings misrepresent consumers’ rights to use copyrighted works. The FTC did not take formal action, but wrote: “the general issue of representations to consumers about the scope of their rights to use copies of works they purchase is of growing importance in light of the widespread and expanding distribution of copyrighted materials in digital form,” and that these issues “will likely…tak[e] on added significance.”

Today, publishers are using similar tactics to attempt to bar AI training on their content, even for fair uses. Penguin Random House (PRH) now includes language in its copyright statement that reads: “No part of this book may be used or reproduced in any manner for the purpose of training artificial intelligence technologies or systems.” The PRH warning “expressly reserves [the titles] from the text and data mining exception” in accordance with the EU copyright directive. Elsevier added a copyright notice to its website that reads, “Copyright © 2025 Elsevier, its licensors, and contributors. All rights are reserved, including those for text and data mining, AI training, and similar technologies.”

When faced with copyright warnings in the wild, it’s critical to remember that misleading warnings or restrictions cannot override fair use and other rights.

Discussion and a Way Forward

This year, we anticipate guidance from the US Copyright Office that will address the legal implications of training AI models on copyrighted works and licensing considerations as part of its study on copyright and artificial intelligence. In a report released this year, the Copyright Office concluded that copyrightability issues related to AI can be resolved under existing law without the need for new legislation. This aligns with Library Copyright Alliance principles on AI and copyright, which hold that the existing US Copyright Act, as applied and interpreted by the Copyright Office and the courts, is fully capable at this time of addressing the intersection of copyright and AI without amendment. Recently, LCA asked the Administration to preserve fair use and protect against contractual override in comments to the White House.

Copyright law in the EU protects the right to conduct text and data mining and AI research for the purposes of scientific research and prohibits private parties from overriding those rights. Although no such provision exists in the US, publishers could begin to level the playing field and simplify the negotiation process by using similar contract language across their international markets—especially those publishers that already have to abide by EU law.

During Copyright Week 2025, ARL published e-Resource Licensing Explained: An A–Z Licensing Guidebook for Libraries, written by Teremi and four other leading licensing librarians with legal experience to help libraries preserve fair use and other rights when negotiating for licensed digital resources.

Libraries can also continue to rely on the advice in the Codes of Best Practices in Fair Use, which include guidance developed by libraries for libraries; other communities have developed their own codes that address fair use in open educational resources, the use of orphan works, documentary filmmaking, scholarly research in communication, software preservation, and more.

Looking ahead to the next two decades, libraries can build on our long history of defending fair use rights to ensure that AI and future technologies can be used to further the objectives of the copyright system: to promote the progress of science and useful arts.