On October 28, 2026, the International Federal of Reproduction Rights Organizations (IFRRO) released a paper entitled “The Legal, Policy, and Market Impact of Artificial Intelligence on Text- and Image-Based Works.” Developed in collaboration with noted copyright expert Professor Eleonora Rosati, the study examines the growing impact of generative AI on authors, publishers and other rightsholders and considers how established copyright principles can address the legal and economic harm arising from the development and use of generative AI. I contacted IFRRO’s CEO Anita Huss-Ekerhult for background.
Many Scholarly Kitchen readers do not know about IFRRO. Before we get into the study, can you give us the elevator pitch?
IFRRO is the global industry body for collective management organizations in the text and image sectors. We engage regularly with governmental and intergovernmental organizations worldwide and facilitate collective management of copyrighted in works through the cooperation of our 160+ members from more than 90 countries. Our members in turn represent millions of creators such as authors, visual artists, and publishers of books, journals, newspapers, magazines and printed music.
Why did IFRRO commission this study?
The rapid development of AI has opened new possibilities, but has also created significant challenges, for authors, publishers and other rightsholders. We all know by now that copyright protected works have, in many instances, been used for AI training without authorization or remuneration. At the same time, AI-generated outputs are having increasingly visible effects on creative markets, including lost clients, declining revenues and layoffs.
While our members feel these impacts, in conversation with regulators and others, we saw a need for evidence. We wanted to understand what “harm” actually means in copyright law and how that concept applies in the context of AI. The study looks at harm from three connected perspectives: legal, policy and market.
We know copies are made and stored throughout the AI process. We wanted to explore whether and how the unauthorized might cause undue harm to the rightsholder. Where permitted under a copyright exception, we wanted to see if it violates international obligations under the Berne three-step test.
The study concludes there is significant damage. What are the primary drivers?
The survey provides a clear picture. A combined 92% of respondents agreed or strongly agreed that AI has already directly and negatively affected earnings.
The examples they provided point to several interconnected forms of harm. One is the use of protected content for AI training without consent and remuneration. Another is the flooding of the market with AI-generated content which reduces the discoverability and profitability of human-created works. And then there are very concrete economic consequences: declining and lost revenues, lost clients, missed professional opportunities and layoffs.
These findings also reflect the broader evidence examined in the study. Across different creative sectors, existing studies point to measurable and, in some cases, already quantified economic and commercial harm. At the same time, a market for AI licensing is already an established and growing commercial reality and this is continually undermined by unauthorized use.
The study also considers some of the latest developments in case law concerning AI and copyright. What do these cases tell us about how existing copyright principles apply in the context of AI?
Yes, and this is an important part of the study. AI may be technologically new, but the copyright principles that courts are being asked to apply are well established. The emerging case law increasingly shows the relevance of questions such as lawful access, the purpose and commercial character of the use, market substitution and, ultimately, whether the use causes undue harm to rightsholders.
To take a recent example from the US, in the appellate decision in Thomson Reuters v. ROSS Intelligence, the court found that the fourth fair use factor weighed against the defendant because its product was meant to be a market substitute for the plaintiff’s product and undercut potential future licensing.
This is significant because it illustrates one of the central points of our study: when assessing market harm, courts may consider not only actual market substitution, but also potential substitution and the impact of an unauthorized use on existing or developing licensing markets.
Interestingly, the survey points to a bizarre secondary effect: a “witch hunt” against genuine human authors. Can you elaborate on that?
Yes, this was a highly concerning trend raised by many authors in the survey. Because the market is flooded with AI content, readers and publishers are highly suspicious. However, because AI detection tools are notoriously unreliable, genuine human writers are being wrongfully accused of using AI simply for using standard literary devices. It is creating an environment of fear, where authors are forced to defend their integrity against unprovable accusations, adding reputational anxiety to their financial stress.
Let’s talk about the law. An overwhelming 96% of your respondents feel the current legal framework is inadequate to protect them against AI developers. What are the specific loopholes or legal interpretations that are causing the most frustration?
The survey identified three principal concerns. First, respondents perceive a lack of laws and policy initiatives adequately addressing the specific challenges raised by AI, including at the international level. Second, they point to inadequate enforcement and sanctions within existing legal systems. Third, they are concerned about interpretations of existing laws that may operate to the undue detriment of copyright protection and in favor of AI developers.
There is deep skepticism regarding how existing laws are being interpreted to favor tech companies over rightsholders. For example, in Europe, members criticized the opt-out mechanism in Article 4(3) of the DSM Directive. In the US, AI developers are trying to aggressively stretch the boundaries of “fair use.”
At the same time, the study makes clear that existing copyright law already contains important safeguards. The three-step test, fair dealing and fair use all require consideration, in different ways, of the effects of unauthorized uses on rightsholders and their markets. So this is not simply a question of creating more law. It is also about ensuring that existing copyright principles are properly interpreted, applied and effectively enforced, and that rightsholders have the transparency necessary to exercise their rights.
Often with copyright, I believe there are adequate laws in place but enforcement is expensive, especially against deep pocketed defendants. How much of the challenge is inadequacy of the laws and how much is simply the cost of litigation?
While strong laws exist, enforcement costs remain a massive barrier for authors and publishers against well-funded defendants, meaning the challenge is equally about the financial impossibility of litigation. There is cause for optimism however. Recent legislative proposals in France [ French Darcos Bill] and Belgium, would introduce legal presumptions that would shift the burden of proof, making enforcement significantly more accessible and affordable for rightsholders.
More positively, a majority of your members believe that AI-specific licenses will greatly improve their economic situation, and we have already seen collective management organizations (CMOs) launch initiatives, including CCC (where I work) in the US, the CLA in the UK, VG Wort in Germany, Tekst&Node in Denmark, JAC in Japan, and Copyright Agency in Australia, to name just a few. How do we ensure these licenses actually protect the “small” authors and publishers?
Licensing is a very important part of the picture. Direct and collective licensing is an established, expanding commercial reality, and collective licensing is historically how the publishing industry scales solutions to technological disruption. We see collective licensing as especially useful in allowing authors and smaller publishers to receive an equitable share of the remuneration.
The broader principle is one of consent. Authors and rightsholders should be in a position to control and meaningfully consent to the use of their content, according to modalities that they are able to define and negotiate.
The study concludes with three non-negotiable recommendations for policymakers and courts to restore a “fair balance”. If you are speaking directly to global regulators, what is your ultimate call to action?
The fundamental message of the study is that situations in which large-scale unauthorized uses of protected content proceed without consent, remuneration or transparency are neither legally allowed nor economically sustainable. They also contribute to the erosion of trust.
The study, therefore, identifies three basic conditions for AI to advance in a way that respects authors’ and rightsholders’ rights:
- Control and Consent Requirement: Rightsholders must have the legal right to control and consent to the use of their content on terms they can meaningfully negotiate.
- Upholding Copyright Principles and Rationale: Policymakers, legislators and courts, must uphold the principles and rationale governing copyright law and avoid that undue harm is caused to authors, publishers and rightsholders.
- Enforceable Transparency: We need legally mandated, enforceable transparency obligations covering both the content ingestion phase and the machine labeling of outputs.
Taken together, consent and appropriate licensing, effective enforcement and meaningful transparency are necessary to ensure a high level of protection for copyright and related rights and to preserve and where necessary, restore the fair balance that the rapid development of AI has unsettled.
