The Spanish government has published a draft Royal Decree aimed at regulating “extended collective licensing” (ECL) for the large-scale use of copyrighted works in the development of general-purpose artificial intelligence (AI). These AI models, particularly Large Language Models (LLMs), are trained on vast datasets, often including texts, images, songs, and other works protected by copyright. Currently, in the United States and the European Union (EU), several lawsuits have been initiated by rights holders against LLM developers for using their works in training these models.
What is the Purpose of the Draft Law?
The objective of the draft law is to address the challenges involved in securing the necessary copyright permissions (reproduction rights) from numerous rights holders whose works are used to train AI models. Obtaining individual authorizations from all rights holders is often impractical and prohibitively expensive.
The draft proposes leveraging “extended collective licensing” as outlined in Article 163 of Spain’s Intellectual Property Law. This mechanism, established under the EU Directive (2019/790) on Copyright in the Digital Single Market (DSM), would enable collective management organizations (CMOs) to grant licenses on behalf of all rights holders, even without individual consent, under specific conditions.
This regulation seeks to ensure fair treatment for all rights holders whose works are used in training, provide an opt-out mechanism, and implement transparent processes. The proposal also specifies requirements for obtaining a representativeness certificate (for CMOs) and obligations for both CMOs and AI developers.
Balancing Rights and Innovation
The proposed regulation acknowledges the innovative potential of AI models while addressing the challenges they pose to creators’ rights. Currently, various mechanisms allow AI developers to use third-party works, including:
- Exceptions under the DSM Directive for “Text and Data Mining” (TDM): Article 3 for scientific research. Article 4 for general use.
- Open licenses such as Creative Commons or Open Database License.
- Access to public sector data.
The Article 4 exception (TDM for industry) is subject to a reservation of rights by rights holders in mined works published online. That is, the rights holder can prohibit TDM on their website or publication. Courts in Germany, for example, suggest such a prohibition is legally valid if it is “effective,” such as through clearly stated legal terms.
The Spanish government’s proposal seeks to establish a licensing mechanism to enable AI developers to obtain rights for protected works available online where rights are reserved, as well as for works not publicly available, such as books, magazines, music, and films in private collections. The ECL system would allow CMOs to grant non-exclusive licenses for the “massive” use of copyrighted works in AI development, even if all rights holders have not explicitly authorized the organization to do so.
Rights of Copyright Holders
The draft recognizes that some rights holders may oppose the use of their works for AI development. These individuals can opt out of the system and prevent CMOs from licensing their rights in this manner. The proposed regulation provides a purportedly “clear” opt-out mechanism, allowing rights holders to retain control over their works. They may:
- Prohibit their use for AI training altogether.
- Allow usage under open licenses such as Creative Commons without royalties or fees.
Non-member rights holders can also approach CMOs to claim royalties collected through these licenses. The draft includes transparency measures to ensure that rights holders are informed about the licensing system, including its scope, duration, revenue-sharing procedures, and opt-out processes.
On the other hand, the draft imposes obligations on AI developers. Those obtaining licenses through this system must respect opt-out choices by rights holders, ensure excluded works are not used, and comply with a license duration limited to three years.
Commentary
The government argues that extended collective licensing is the optimal solution to address the challenges of licensing copyrighted works for AI development. It aims to minimize administrative burdens and provide legal clarity.
However, this mechanism shifts the default value of accessing published works for TDM (legitimate under many circumstances per the DSM Directive) from “use unless explicitly restricted by the rights holder” (favoring innovation with publicly available works) to “obtain a license unless the rights holder has opted out.” This raises significant concerns, particularly for creators of works under open licenses like Creative Commons (over 2.5 billion works online), who would need to opt out of the default licensing framework.
Additionally, CMOs must establish fees for the effective use of works and justify the pricing of licenses for training data. Determining value—e.g., whether an image, text, or song is worth more, and for what type of AI and purpose—remains an unresolved complexity.
The draft aims to create a framework that fosters AI innovation while safeguarding creators’ interests. By simplifying licensing for mass data usage, it seeks to ensure fair remuneration for rights holders and a clear mechanism for those opting out. However, the proposal might incentivize AI developers to migrate to countries with more favorable access to training data, such as Japan or—depending on ongoing litigation outcomes—the United States.
At Across Legal, we are already advising several business and scientific research projects on data usage for AI training, the impact of open licenses, and the negotiation of specific dataset access licenses. For further insights, contact our IT and IP Department. Additionally, we recommend reading our colleagues’ commentary on the Lucentinvs blog, among others.
