On a significant Friday afternoon, the U.S. Copyright Office unveiled a comprehensive report that delves into the intricate relationship between copyrights and the burgeoning field of generative AI training. This report notably supports the necessity of licensing copyrights when such materials are utilized in commercial AI training, indicating a pivotal move towards balancing intellectual property rights with technological advancement.
On the subsequent Saturday, May 10, the nation’s foremost copyright authority, Shira Perlmutter, who held the position of Register of Copyrights, was unexpectedly terminated by President Donald Trump. This dismissal closely followed the firing of Carla Hayden, the Librarian of Congress who appointed and oversaw Perlmutter. In the aftermath, Rep. Joe Morelle (D-NY), a member of the House Administration Committee responsible for the oversight of the Copyright Office and the Library of Congress, suggested that the timing of these dismissals was not coincidental. He emphasized that it occurred less than a day after the Copyright Office declined to endorse Elon Musk’s controversial initiative to utilize extensive copyrighted works for training AI models.
The report has been widely interpreted as a favorable outcome for copyright owners within the music industry. It articulates three pivotal positions: the Copyright Office’s endorsement of licensing copyrighted materials when a “commercial” AI model employs them for training, its rejection of compulsory licensing as an appropriate framework for future licensing, and its dismissal of any “opt-out approach” to copyright usage. These positions highlight the Office’s commitment to protecting the rights of creators while fostering innovation in AI.
The Copyright Office asserts that in cases deemed “commercial,” licensing copyrights for AI training presents a “practical solution.” It further states that utilizing copyrighted materials without proper licensing extends “[go] beyond established fair use boundaries.” The report underscores that certain commercial AI models are in direct competition with existing market offerings, which raises significant concerns regarding copyright infringement. Nevertheless, the Office clarifies that if an AI model is developed for “purposes such as analysis or research”—essential activities vital for maintaining international competitiveness—then its outputs are unlikely to serve as substitutes for the original works that informed their training.
“In our perspective, fostering American leadership in the AI industry would be best achieved by supporting both of these world-class sectors that contribute immensely to our economic and cultural progress. Implementing effective licensing options can ensure that innovation continues to thrive without compromising intellectual property rights,” the report articulates. This statement reflects a balanced approach towards nurturing creativity and safeguarding the interests of copyright holders.
While the report advocates for collaborative licensing between copyright holders and AI companies, it acknowledges that the majority of stakeholders are not in favor of any statutory changes or “government intervention” in this domain. “The Office believes…[that] would be premature at this time,” the report states. It further mentions, “we concur with commentators that a compulsory licensing framework for AI training could present significant drawbacks. Such a framework would establish fixed royalty rates and terms that could become entrenched within the industry, making them challenging to amend. Additionally, premature adoption risks stifling the development of flexible and inventive market-based solutions. Compulsory licenses may take years to establish, often necessitating tedious negotiations on numerous operational specifics.”
The Copyright Office also considers the viewpoints of music industry organizations, including the National Music Publishers’ Association (NMPA), American Association of Independent Music (A2IM), and the Recording Industry Association of America (RIAA). These organizations collectively express strong opposition to the concept of a future compulsory or government-controlled licensing system for AI training. The music sector currently grapples with a compulsory licensing structure for mechanical royalties, which allows the government to regulate rates for one category of royalties accrued from streaming and sales.
“The majority of respondents addressing this matter voiced their opposition or raised concerns about the potential for compulsory licensing,” the report notes. “Representatives of copyright holders and creators contend that the compulsory licensing of their works for use in AI training would undermine their ability to control the utilization of their creations, asserting that no market failure exists to justify such measures. A2IM and RIAA warned that compulsory licensing could lead to ‘below-market royalty rates, additional administrative burdens, and… restrictions on innovation.’ Moreover, NMPA characterized it as ‘an extreme remedy that strips copyright owners of their right to freely negotiate in the market, depriving them of the ability to choose their business partners, dictate how their works are utilized, and determine their compensation.’”
The Copyright Office delegates the responsibility of determining appropriate licensing and compensation methods for training data to the copyright owners and AI companies. However, it does present several potential options for consideration. These options include “compensation structures based on a percentage of revenue or profits.” Nevertheless, if the free market fails to establish a suitable licensing solution, the report suggests that “targeted intervention such as [Extended Collective Licensing] ECL should be contemplated.”
ECL, employed in several European nations, would permit a collective management organization (CMO) to issue and oversee blanket licenses for “all copyrighted works within a specific category.” This system is reminiscent of existing practices in the music industry, where organizations like The MLC (The Mechanical Licensing Collective) and performing rights organizations (PROs) such as ASCAP and BMI operate. However, the key distinction of an ECL system is that it allows the CMO to license works for individuals who have not actively joined the organization. Although these ECL licenses are still negotiated within a “free market,” the government would “regulate[e] the overall system and exercise[e] a degree of oversight.”
Despite some apprehensions from AI firms that blanket licensing by copyright holders could lead to antitrust complications, the Copyright Office has sided with copyright holders, asserting that “[the] courts have determined that there is nothing inherently anticompetitive about collective or even blanket licensing of copyrighted works, provided that certain safeguards are put in place—such as ensuring that licensees maintain the option to obtain direct licenses from copyright owners as an alternative.”
This is a “pre-publication” version of an anticipated final report, which is expected to be released in the “near future without any substantive alterations anticipated,” according to the Copyright Office. The Office noted that this “pre-publication” version was expedited to address inquiries from Congress and key stakeholders.
This report signifies the Office’s third examination of generative AI and its ramifications on copyrights since it initiated an inquiry into this matter in 2023. The inaugural report, issued on July 31, 2024, concentrated on the theme of digital replicas. The second report, published on January 29, 2025, focused on the copyrightability of outputs generated by generative AI.
