EU Parliament Report Declares GenAI Training Data Use Not 'Fair Use,' Advocates for Creator Compensation

The European Stance: Why GenAI Training Data Isn't 'Fair Use' Under EU Law

The rapid proliferation of generative artificial intelligence (GenAI) has ignited a fierce global debate surrounding intellectual property rights, particularly the use of vast quantities of copyrighted material to train these powerful models. At the heart of this controversy lies the question of whether such training constitutes 'fair use' or falls under existing copyright exceptions. A recent research paper, commissioned by the European Parliament, delivers a clear and impactful message: under current EU law, the 'fair use' defense for training GenAI models on copyrighted content simply 'does not compute'.

This comprehensive study, requested by the Committee on Legal Affairs and conducted by the Policy Department for Justice, Civil Liberties and Institutional Affairs, delves deep into the complexities of applying existing copyright frameworks to the unique nature of generative AI. Its findings challenge the core arguments put forth by many AI developers and companies, who have often likened the process of training an AI model to a human student learning from books or artworks – an activity typically covered by fair use or similar exceptions.

Deconstructing the 'Human Learning' Analogy

One of the central tenets of the report is its direct refutation of the analogy between human learning and AI training. While superficially appealing, the study argues that this comparison is fundamentally misleading from a legal standpoint within the context of EU copyright law. When a human reads a book or studies a painting, they internalize concepts, styles, and knowledge, but they do not typically make literal copies of the entire work for the purpose of creating new, potentially competing expressive content on a massive scale.

The report emphasizes that generative AI models, during their training phase, typically involve the creation of numerous temporary or permanent copies of the copyrighted works. They process the 'actual expressions' found within these works, not just the underlying ideas or facts. This process, according to the study, transcends the boundaries of what is permissible under current EU legal exceptions designed for activities like research or analysis, such as the text-and-data mining (TDM) exception.

Philosopher Luciano Floridi's perspective, cited in the report, further underscores this distinction: AI 'acts without understanding'. It operates by identifying and replicating statistical patterns and correlations within the training data, rather than engaging with the semantic meaning or creative intent in the way a human consciousness does. This fundamental difference in how AI processes information is deemed legally significant, rendering the 'student reading a book' analogy inappropriate for justifying the use of copyrighted material in commercial GenAI training.

The Inadequacy of the Text-and-Data Mining Exception

The EU's legal framework includes a text-and-data mining (TDM) exception, primarily introduced to facilitate scientific research and data analysis. However, the European Parliament-commissioned study concludes that this existing exception was 'not designed to accommodate the expressive and synthetic nature of generative AI training'. Applying the TDM exception to GenAI systems, the report warns, 'risks distorting the purpose and limits of EU copyright exceptions'.

The TDM exception typically allows for the automated analysis of large amounts of text and data in digital form to generate information, such as patterns, trends, and correlations. While AI training involves analyzing data, the *purpose* and *outcome* of generative AI training differ significantly from traditional TDM. GenAI is trained to *generate* new content that is often expressive and directly competes with the original works used for training. This generative aspect, the report argues, falls outside the intended scope of the TDM exception, which is generally aimed at extracting non-expressive information or insights.

Furthermore, the TDM exception often includes provisions allowing rightsholders to opt-out of having their works used for TDM. While some AI companies have argued that this opt-out mechanism is sufficient, critics and the new report suggest that placing the burden of opting out on individual creators is impractical and fails to address the fundamental issue of unauthorized commercial use for generating competing content. The sheer scale of data scraping required for training large models makes effective opt-out mechanisms challenging to implement and enforce for millions of creators.

A Call for New Legislation and Equitable Remuneration

Given the perceived shortcomings of the existing legal framework, the report does not merely criticize the status quo; it proposes concrete legislative action at the EU level. The study advocates for the introduction of a new, specific EU-level statutory exception tailored precisely for the purpose of training generative AI systems. This new exception would acknowledge the unique technical requirements of AI training while simultaneously establishing clear boundaries and obligations.

Crucially, the report couples this proposed exception with the introduction of an 'unwaivable right to equitable remuneration for authors and rightsholders whose works are used in such training'. This is a significant recommendation, suggesting that creators should have a legal right to be compensated when their copyrighted works contribute to the training data of commercial GenAI models, regardless of any opt-out clauses or interpretations of fair use.

The concept of 'equitable remuneration' implies a system where compensation is distributed fairly among rightsholders whose works are utilized. The practical implementation of such a system would undoubtedly be complex, requiring mechanisms for identifying which works were used in training data (a challenge given the opaque nature of many training datasets) and establishing fair compensation models. However, the report signals a clear intent to ensure that the value derived by AI developers from creative works is shared with the original creators.

Addressing AI-Assisted Works

Beyond the training data issue, the report also touches upon the copyright status of content produced by AI systems. It argues that while outputs generated entirely by machines, without human creative input, should likely remain unprotected by copyright, works created with the assistance of AI require 'harmonised protection criteria'.

This distinction is vital. An image or text generated solely by an AI model based on a simple prompt might not meet the threshold of originality required for copyright protection in many jurisdictions. However, a work where a human creator uses AI tools as instruments to enhance, modify, or generate elements of their creation, exercising their own creative choices and direction, could potentially be eligible for copyright. The report suggests the need for consistent rules across the EU to determine when and how such AI-assisted works qualify for protection, likely focusing on the degree and nature of human creative involvement.

Parallels and Contrasts with the US Landscape

The European Parliament's commissioned report echoes sentiments and legal challenges that have been gaining momentum across the Atlantic, particularly in the United States. The US legal system employs the doctrine of 'fair use', a flexible defense that allows limited use of copyrighted material without permission for purposes such as criticism, comment, news reporting, teaching, scholarship, or research. However, the application of fair use to the large-scale commercial training of GenAI has become a major point of contention.

In May, the head of the US Copyright Office, Shira Perlmutter, was reportedly removed from her position following her agency's conclusion that the use of copyrighted material by AI developers often exceeded the bounds of existing fair use doctrine. The US Copyright Office had indicated that fair use does not typically cover the commercial exploitation of vast volumes of copyrighted works to produce new expressive content that directly competes in existing markets. This stance aligns closely with the perspective articulated in the new EU report.

The US is currently a hotbed of litigation challenging the use of copyrighted data for AI training. Numerous lawsuits have been filed by artists, writers, programmers, and media organizations against prominent AI companies. For instance, major studios like Disney and Universal have initiated legal action against GenAI providers such as Midjourney. Their claim alleges that Midjourney's training data includes copyrighted characters and works from their extensive catalogs, constituting a form of 'plagiarism' on a massive scale.

These US lawsuits, alongside the stance taken by the US Copyright Office, indicate a growing international consensus among rightsholders and some regulatory bodies that the current application of 'fair use' or similar exceptions is insufficient to address the economic and legal implications of using copyrighted content for commercial GenAI training. The European Parliament report adds significant weight to this perspective from a major global regulatory power.

The debate over AI and copyright is not confined to legal interpretations; it has tangible economic consequences for creators and creative industries. Writers, musicians, visual artists, and other content producers rely on copyright to protect their livelihoods and control how their work is used. The widespread, uncompensated use of their work in AI training datasets raises concerns about devaluing creative output, reducing opportunities for human creators, and disrupting established licensing markets.

The call for 'equitable remuneration' in the EU report is a direct response to these economic concerns. It seeks to establish a mechanism whereby creators can receive a share of the economic value generated by AI models that are built upon their creative legacy. This could potentially involve collective licensing schemes, statutory levies, or other models for distributing royalties.

Challenges and the Path Forward

Implementing the recommendations of the European Parliament report will not be without its challenges. Defining the scope of a new statutory exception for GenAI training, establishing a fair and effective system for equitable remuneration, and harmonizing criteria for AI-assisted works across all EU member states are complex tasks. Technical hurdles exist in tracing the origin of training data and attributing usage to specific rightsholders.

However, the report signals a clear direction for future EU policy. It suggests that the European Union is likely to pursue a regulatory path that prioritizes the rights and compensation of creators in the age of generative AI. This could involve amendments to existing copyright directives or the introduction of new legislation specifically addressing AI and intellectual property.

The report's findings will likely inform discussions within the European Parliament and the European Commission as they continue to grapple with the broader implications of artificial intelligence, including those addressed in the comprehensive EU AI Act. While the AI Act primarily focuses on the safety and ethical aspects of AI systems, the copyright dimension remains a critical piece of the regulatory puzzle.

The global nature of AI development and the internet means that regulatory approaches in major jurisdictions like the EU and the US will influence each other and set precedents for other countries. The alignment between the stance taken in the EU report and recent developments in the US suggests a potential convergence towards stricter interpretations of copyright exceptions for AI training and a greater emphasis on compensating creators.

Ultimately, the European Parliament-commissioned report serves as a significant intervention in the ongoing debate about AI and copyright. By firmly stating that current EU law does not support the 'fair use' defense for commercial GenAI training and advocating for mandatory creator compensation, it sets the stage for potential legislative changes that could reshape the relationship between AI developers and the creative industries, aiming to ensure that the benefits of the AI revolution are shared more equitably.

The path forward involves navigating complex legal, technical, and economic considerations. Policymakers will need to balance the need to protect creators' rights and incentivize creative output with the desire to foster innovation in the field of artificial intelligence. The report provides a strong foundation for the argument that this balance must include fair compensation for the creative works that serve as the building blocks for the next generation of AI.

The implications of this report extend beyond legal theory. For developers, it signals the potential need for new business models that incorporate licensing or compensation mechanisms for training data. For creators, it offers hope for a legal framework that recognizes the value of their contributions in the AI era. As the EU considers its next steps, the principle that using copyrighted work for commercial GenAI training requires proper authorization and remuneration appears to be gaining significant traction.

This development is part of a larger global conversation about how to adapt existing legal frameworks, designed for a pre-AI world, to the realities of advanced machine learning. The report's clear stance against the 'fair use' argument for training data and its call for equitable remuneration mark a pivotal moment in this ongoing dialogue, potentially paving the way for new standards in AI and intellectual property law.

Key Takeaways from the Report

The European Parliament-commissioned report highlights several critical points:

Current EU copyright exceptions, including Text and Data Mining (TDM), are inadequate for generative AI training.
The analogy between human learning and AI training is legally misleading under EU copyright law.
AI training involves copying and processing the 'actual expressions' of copyrighted works, exceeding the scope of existing exceptions.
A new EU-level statutory exception for GenAI training is needed.
This new exception should include an unwaivable right to equitable remuneration for rightsholders.
Fully machine-generated outputs should remain unprotected by copyright.
AI-assisted works require harmonized protection criteria based on human creative input.

These points collectively form a strong argument for legislative intervention to address the copyright implications of generative AI training data use in the European Union.

Global Context and Future Implications

The EU report does not exist in a vacuum. It is part of a global trend where policymakers and courts are grappling with the disruptive impact of AI on copyright. The legal challenges in the US, such as the lawsuits brought by artists and companies against AI developers, underscore the urgency of finding solutions.

For instance, the lawsuit filed by Disney and Universal against Midjourney is a high-profile example of copyright holders asserting their rights against AI companies they accuse of unauthorized use of their creative assets for training. Such cases test the boundaries of existing copyright law and fair use doctrines in court.

Beyond litigation, various stakeholders are exploring potential solutions, including licensing frameworks, collective rights management organizations for AI training data, and technological measures to identify and track the use of copyrighted works. Cloudflare, for example, has proposed concepts like an 'AI crawler tollbooth' to allow publishers to control and potentially monetize access to their content for AI training purposes.

The European Parliament report's recommendation for an 'unwaivable right to equitable remuneration' suggests a preference for a system that guarantees compensation to creators, potentially through collective management, rather than relying solely on individual licensing deals or opt-out mechanisms.

As the EU moves forward, the recommendations of this report will likely play a crucial role in shaping its legislative response to the challenges posed by generative AI. The outcome will have significant implications not only for the European creative industries and AI sector but also for the global development of AI and intellectual property law.

The debate is far from settled, but the European Parliament-commissioned report marks a clear step towards a regulatory environment that seeks to ensure creators are fairly compensated when their work fuels the AI revolution. It reinforces the idea that while AI may 'learn' from human creativity, that learning process, when commercialized, should respect and reward the original sources.

The principle that underpins the report's findings is that the economic benefits derived from training AI models on copyrighted works should not accrue solely to the AI developers. Creators, whose work forms the essential foundation of these models' capabilities, have a legitimate claim to a share of that value. Establishing a legal right to equitable remuneration is presented as a necessary step to maintain a vibrant creative ecosystem in the age of artificial intelligence.

The report's emphasis on the legal distinction between human learning and AI training provides a strong theoretical basis for its conclusions. By framing AI training as a technical process involving the copying and processing of expressive content, rather than a mere intellectual absorption of ideas, it positions this activity firmly within the realm of copyright regulation.

The call for harmonized criteria for AI-assisted works also addresses a growing need for clarity. As AI tools become more integrated into creative workflows, determining authorship and copyright ownership becomes increasingly complex. Clear guidelines are essential for both creators using AI and for the legal system to handle potential disputes.

In conclusion, the European Parliament-commissioned report is a landmark document that provides a robust legal analysis challenging the 'fair use' defense for GenAI training data in the EU. Its recommendations for a new statutory exception and mandatory equitable remuneration signal a potential shift towards a more protective and compensatory legal framework for creators in the face of advancing artificial intelligence technology. The global implications of this stance are significant, contributing to an international conversation about how to balance technological innovation with the fundamental rights of intellectual property holders.

The report serves as a powerful reminder that as AI capabilities expand, so too must our legal and ethical frameworks evolve to ensure fairness and sustainability for the creative industries that provide the very fuel for AI's generative power.

Subscribe to Our Tech & Career Digest

EU Parliament Report Declares GenAI Training Data Use Not 'Fair Use,' Advocates for Creator Compensation