The European Parliament adopted a Resolution on Copyright and Generative Artificial Intelligence on March 10, 2026, that signals a potentially significant shift in European copyright policy. Although the Resolution itself is not legally binding, it reflects the growing view within EU institutions that traditional copyright rules may not adequately address the realities of generative AI training. In particular, the Committee Report underlying the Resolution suggests a broader territorial reach of EU copyright law for AI systems offered in the European market. It suggests a potentially retroactive flat rate licensing fee "5 to 7% of [the] global turnover” to compensate the creative industry for the training. If implemented in future legislation, this approach could affect AI developers and deployers worldwide, including companies based in the United States. It may lead to tensions with non-EU legal regimes governing AI training practices.
On March 10, 2026, the European Parliament adopted a Resolution entitled “Copyright and Generative Artificial Intelligence – Opportunities and Challenges.” The Resolution builds on a report prepared by the Parliament’s Committee on Legal Affairs (JURI), which examines the interaction between traditional copyright law and the rapid development of generative artificial intelligence. Although the Resolution itself does not create binding law, it reflects a political signal from the European legislature: the current copyright framework may no longer be sufficient to address the realities of AI training.
For companies operating in the AI ecosystem, developers of foundation models, providers of AI-enabled products, and businesses integrating generative AI into their operations, the debate in Europe is highly relevant. The policy direction emerging in Europe could significantly affect how training data are sourced, documented, and whether licenses from the creators are needed. Perhaps most importantly, the discussion suggests that the EU may be prepared to interpret the territorial reach of copyright law more broadly in the context of AI systems, which could lead to potential tensions with legal regimes outside the EU.
The Committee Report and the Resolution contain an number of elements that are entirely new for EU copyrights, The reasons why the parliamentarians believe that a reform for AI training data sets is necessary and the main elements of their proposals can be summarized as follows:
AI Training and the Limits of the Existing EU Copyright
- From printing presses to machine learning - European copyright law historically developed in a technological environment very different from today’s digital economy. Its central objective has been to protect individual creative works against unauthorized acts of reproduction and distribution. The legal framework reflects a world in which copying was easy to identify and control. Generative AI systems operate in a fundamentally different manner. Rather than interacting with individual works, they process extremely large datasets that may contain billions of digital copies of texts, images, audio files, and videos. These systems identify patterns within those datasets, derive statistical relationships, and generate new outputs based on that analysis. From a technological perspective, this process differs substantially from the classical forms of copyright exploitation. AI training generally does not involve the commercial distribution of the underlying works themselves. Instead, it relies on large-scale computational analysis of data that may include copyrighted material.
- The growing economic importance of generative AI - Generative AI has already become an integral component of modern business processes. Companies rely on these technologies to produce text, software code, visual content, and analytical insights. Behind this technological capability lies a legally sensitive question: what content was used to train the AI model? In many cases, the training includes large amounts of journalistic articles, photographs, music, books, and other creative works produced. These works are typically protected under European copyright law. Their use as training data therefore raises fundamental legal questions about authorization, transparency, and compensation. A central legal debate in EU policy discussions is whether the text-and-data-mining (TDM) exception in the 2019 EU Copyright in the Digital Single Market Directive (CDSM Directive) covers generative AI training. Many commentators and industry groups argue that the exception was never designed for large-scale AI model training.
- A policy conclusion emerging in Europe - The Report prepared by the JURI Committee arrives at a notable conclusion: the traditional copyright framework may no longer be sufficient as the sole regulatory structure for the AI era. The Report therefore raises the possibility that additional legal mechanisms may be required to balance technological innovation with the protection of creative works. This debate reflects a broader policy tension within Europe. On the one hand, the European Union seeks to strengthen its competitiveness in artificial intelligence and reduce the technological gap with the United States and China. Achieving this objective requires powerful AI models and access to large quantities of high-quality training data. On the other hand, media companies, publishers, musicians, filmmakers, and digital platforms produce much of the content for Generative AI. Many stakeholders in the creative industries argue that their works are used without authorization and without adequate compensation. Accordingly, the Report and the Resolution reflect two simultaneous policy objectives: regulating AI training and protecting the economic sustainability of Europe’s creative industries. Specifically The Explanatory of the Report suggests: “[…] the European legislator or the European Commission should, […] establish an immediate, simple, flat-rate copyright fee for this use of 5 to 7% of global turnover in order to compensate for the added value that these businesses generate using the data of European creatives and to ensure it remains in Europe.”
Key Elements of the European Parliament’s Approach
- Legal uncertainty illustrated by court proceedings - The legal uncertainty surrounding AI training is already visible in national court proceedings. In November 2025, the Regional Court of Munich issued a decision in litigation brought by the German collecting society GEMA against OpenAI. The court concluded that certain forms of AI training involving protected works could constitute copyright infringement and that the EU text-and-data-mining exception did not apply in the circumstances of the case. The decision is not yet final. The outcome contrasts with litigation in the United Kingdom also of last November. In the case Getty Images v. Stability AI, the High Court in London dismissed Getty’s copyright infringement claim, concluding that the Stable Diffusion model could not be considered an infringing copy under the applicable legal standard. The diverging outcomes illustrate that courts across European jurisdictions are still grappling with how traditional copyright concepts should apply to generative AI technologies.
- Opt-out mechanisms and a potential EUIPO registry - One proposal discussed in the JURI Report and now in the Resolution is the creation of standardized opt-out mechanisms allowing rights holders to signal that their content should not be used for AI training. The Report and the Resolution suggest that such signals should be machine-readable so that AI developers can automatically recognize them during the data collection process. To facilitate implementation, the explanatory statement accompanying the Resolution proposes that opt-out declarations could be recorded in a centralized European register maintained by the European Union Intellectual Property Office (EUIPO). Under such a system, AI providers would likely be forced to consult a future EUIPO register to determine whether specific works are excluded from AI training.
- Increased transparency requirements - Another central theme of the Report is transparency. The European Parliament emphasizes that developers of generative AI systems should disclose more information than already required by existing law about the copyrighted materials used in training their models. Some scenarios discussed in the Resolution envision that this information could be communicated to a public authority such as EUIPO, which could then act as a point of contact for rights holders and play a role in monitoring compliance.
- Enforcement and evidentiary rules - The Report also discusses the possibility of introducing a rebuttable presumption in copyright enforcement. If AI providers fail to comply with transparency obligations, it could be presumed that relevant works were used in training unless the developer proves otherwise. Arguably, such a rule would significantly strengthen the position of rights holders in litigation by lowering evidentiary barriers for copyright claims.
A Potential Expansion of the Territorial Reach of EU Copyright Law
- The traditional territoriality principle - Copyright law has traditionally been governed by the territoriality principle. Under this concept, copyright protection is determined by the laws of each jurisdiction, and acts occurring within a given territory are governed by that territory’s law. This principle has historically allowed different legal systems to coexist even when they adopt different approaches to copyright protection.
- The approach suggested in the Resolution - The Resolution suggests that this traditional understanding of territoriality may need to be reconsidered in the context of AI: EU copyright rules should apply whenever an AI system is offered or used within the EU internal market, regardless of where the underlying model was trained. In practical terms, this means that an AI model trained in the United States or Asia could still be subject to EU copyright requirements if it is made available to users in Europe. The objective is to ensure a level playing field between European and non-European AI providers operating in the EU market.
- Potential implications for global technology companies - For companies outside Europe, this development could have far-reaching implications. If the EU moves toward applying its copyright rules based on market access rather than the physical location of training activities, AI providers could become subject to EU regulatory requirements even when development takes place entirely outside the Union. Such an approach could create tensions with legal regimes in other jurisdictions, particularly where those regimes take a different view on the legality of AI training practices.
Practical Implications for U.S. Companies
The European Parliament’s Resolution does not have direct legal effect. Legislative initiatives in the European Union must be proposed by the European Commission before binding rules can be adopted. Nevertheless, parliamentary Reports and Resolutions frequently provide early indicators of future regulatory developments.
For U.S. technology companies, the policy discussion emerging in Europe suggests that compliance with EU copyright rules may increasingly depend on transparency regarding training data, documentation of data sources, and mechanisms for respecting rights-holder opt-outs. At the same time, the evolving debate on territorial reach indicates that companies outside Europe may still be affected if their AI services are offered to European users.
Businesses that rely on generative AI, whether as developers, service providers, or enterprise users, may therefore wish to assess how their AI systems interact with European copyright rules, how training data sources are documented, and whether governance frameworks adequately address the emerging regulatory landscape in Europe. Generative AI involves a complex “AI supply chain” consisting of dataset providers, AI model developers, application developers, and enterprise users. Each of these actors potentially interacts with copyrighted material in different ways. In practice, this raises questions such as: Is copyright liability primarily with foundation model developers? Could downstream deployers also face copyright exposure? How will contractual allocation of risk evolve?
The European debate around AI training currently intersects with several different legal regimes, including copyright law, the EU AI Act, and data protection law. If the EU ultimately adopts specific copyright-related requirements for AI training, such as transparency obligations, opt-out systems, or licensing mechanisms, these rules may differ significantly from those applied in other jurisdictions.
The debate in Europe is gradually moving beyond classical copyright law toward a broader concept of data governance for AI training. For globally operating companies, the interaction between EU copyright law, emerging AI regulation (such as the EU AI Act), data privacy (GDPR) and differing legal approaches in other jurisdictions may soon become a central issue in cross-border technology compliance. As a result, AI providers and users may increasingly face a situation in which a single AI model must comply with multiple, potentially conflicting regulatory frameworks depending on where it is deployed. Training practices that are considered lawful in one jurisdiction may trigger legal risks in another.
Key EU Documents
Report of the European Parliament (JURI Committee):
https://www.europarl.europa.eu/doceo/document/A-10-2026-0019_EN.html
Explanatory Statement of the JURI Committee:
https://www.europarl.europa.eu/doceo/document/A-10-2026-0019_EN.html#_section2
European Parliament Resolution:
https://www.europarl.europa.eu/doceo/document/TA-10-2026-0066_EN.html
