We have some exciting news. The Kluwer Copyright Blog is going through a makeover! The blog’s layout will be updated from the middle of July. This means that for the first half of July we will not be posting as the systems are being updating. Nonetheless, there is a lot to catch up with – read Sabine Jacques’ posts on the AG Opinion on Pelham II (here and here), as well as João Quintais’ mid-term review of posts on AI/copyright. After the pause, you will get a chance to catch up on our second trimester EU copyright law roundup, as well as reports on the UCL roundtables as part of its genAI and copyright law series.

In the meantime, take this opportunity to recharge and we will see you in a few weeks!

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

Welcome back to Part II of the analysis of AG Emiliou’s Opinion in C-590/23 Pelham II. In Part I (here), we analysed the interpretation of pastiche as an autonomous concept of EU law, along with its defining features.

In this Part II, we turn to the balancing of fundamental rights. Here, the AG ventures into bold territory, challenging assumptions about the scope of copyright protection, emphasising the importance of artistic freedom, and sending a clear signal to both the EU legislator and industry stakeholders. We’ll also explore what’s missing and what this means for the recording industry.

The great in-between freedoms…

The concluding section of AG Emiliou’s opinion highlights the inherent complexity in balancing the fundamental rights of artistic freedom and freedom of expression with the right to property. While these rights are formally recognised as equal under the Charter, the opinion clearly tilts the balance towards the supremacy of artistic freedom.

Crucially, the opinion oversimplifies the extent to which the EU legislature is afforded discretion in mediating between these competing rights. While some cited case law (ECtHR Ashby Donald, paras 40–41; Neptune Distribution, para 76) does suggest that a broader margin of appreciation may be granted where commercial speech is concerned, this does not imply that such a margin should extend entirely to artistic expression. Instead, those judgments seem to support the view that national authorities are often better positioned to undertake the nuanced balancing required in such commercial cases and ECtHR case law does emphasise a narrower margin in case of artistic expression (ECtHR Vereinigung Bildender Künstler v. Austria, noting that in the case of artistic expressions, the margin should be particularly limited; para 6). By asserting a wide legislative margin of appreciation at the EU level, however, the opinion risks undermining the careful balancing role traditionally reserved for national courts and potentially diminishes the protections afforded to artistic freedom, which demands a more stringent and context-sensitive approach. Admittedly, the ECtHR’s own position is not without ambiguity. For instance, in Sousa Goucha v Portugal, the Court acknowledged that particular care must be exercised due to the potential implications for freedom of expression, yet still held that national authorities enjoyed a wide margin of appreciation (para 50), reflecting a certain tension in its reasoning and going against a well-established line of case law.

Central to the AG’s analysis is the recognition of the social function of pastiche as a vehicle for cultural commentary and artistic engagement. Against this backdrop, AG Emiliou suggests that artistic freedom ought, in practice, to take precedence over proprietary interests where the creative reuse of protected content is concerned. Relying on Article 17(2) of the Charter by producers and broadcasters is, in his view, less persuasive when measured against the broader objectives of cultural exchange and creative reuse. These related rights must be interpreted considering their functional purpose and should not be extended so far as to inhibit forms of creative appropriation such as legitimate pastiche (paras 114-115).

The opinion goes further still, suggesting what we can find in relation to databases. It asserts that right-holders should only be protected against the communication of parts of their protected subject matter (such as samples of phonograms) when those parts are sufficiently substantial, either quantitatively or qualitatively, to undermine their ability to secure an adequate return on investment (para 117). This threshold is linked to the risk of market substitution: protection is warranted only where the reuse could generate a competing product capable of adversely affecting sales or other legitimate commercial uses.

Yet, the AG draws a distinction between related and authorial rights. While the balance might shift when dealing with the musical composition itself (e.g., the score or lyrics), the opinion maintains that in the context of related rights, such as those held by phonogram producers, the equilibrium should favour artistic freedom (para 128). However, this delineation raises potential complications. As the AG himself recognises, in cases of sampling, both the recording (master rights) and the underlying composition (authors’ rights) may require clearance. If the goal is to foster legal certainty and facilitate creative reuse, privileging artistic freedom in one domain while maintaining restrictive standards in another may inadvertently entrench complexity, rather than reduce it. The elevation of artistic freedom in this context, while laudable in principle, may therefore produce outcomes that are at odds with the practical realities of music production and licensing.

The silence of the Opinion

It is important to note what is not covered by the opinion and which may need further consideration. There is nothing about the interpretation of the three-step test as prescribed in article 5(5) InfoSoc Directive and derived from international treaties such as first appeared in the Berne Convention, article 9(2) which states that: ‘[…] the reproduction of such [i.e. literary and artistic] works in certain special cases, provided such reproduction does not conflict with the normal exploitation of the work and does not unreasonably prejudice the legitimate interests of the author’. This test is well-rooted in the international copyright framework and can also be found in article 13 TRIPS and article 10(2) WIPO Copyright Treaty. Nevertheless, the opinion does confirm that, in order for an exception to apply, the use must not only satisfy the requirements of a specific exception but also need to conform with the three-step test on a case-by-case basis (para 125). And in relation to other exceptions, the three-step test was used to add a proportionality requirement. This can be seen in the Painer (at 134) and Deckmyn (at 27) decisions whereby the application of the exception is also subject to ‘strike a fair balance between the right to freedom of expression of users of a work or other protected subject-matter and the reproduction right conferred on authors’. Here, the opinion remains vague, leaving one to wonder about the reason for this omission.

Against all obiter

Overall, the AG’s opinion reflects a clear openness to enhancing flexibility within the EU copyright system, particularly by recognising the imperative to better align it with the fundamental right to artistic freedom. Notably, the AG interprets the pastiche exception in a broad manner (although not as broad as some may want it to be), while also directing a pointed message to the EU legislator: more comprehensive legislative reform is needed to modernise copyright law considering Charter values (see especially paras 79 and 113–132). Alongside this, the opinion delivers a strong call to right-holders and collective management organisations to adapt their licensing models and remove barriers to creative reuse (para 100), while also acknowledging the complexities specific to the music industry (paras 101–102).

However, such an approach is unlikely to be welcomed by all. By effectively weakening the traditional exclusive rights under copyright, the opinion challenges long-standing assumptions in the creative industries and markets built on rights granted. Although it commendably centres the position of grassroots creators, it remains unclear whether the interpretation offered would provide them with practical legal certainty or simply add new layers of complexity, as I have noted in relation to master vs composition rights, which may further obscure rather than clarify the legal framework. And while the freedom to reuse creatively is valorised, there is a risk that lesser-known artists whose work is sampled might miss out on potential financial gains should their material be included in a later commercially successful work.

At this point, given the opinion’s otherwise forward-looking tone and its explicit concern with contemporary modes of creative production, one might have expected at least a brief engagement with generative AI and digital replicas, tools that not only rely heavily on pastiche-like techniques but also pose new challenges to the boundaries of copyright. While the mention of Andersen v Stability AI and Getty Images v Stability AI in footnote 103 gravitates toward this area, it is confined to a narrow observation about right-holders seeking protection over style and therefore, justifying the need for a pastiche exception rather than a deeper reflection on how such technologies complicate existing legal concepts. The opinion signals a shift from the traditionally strong protection of right-holders in EU copyright law, subordinating exclusive rights to artistic freedom. This shift may empower grassroots creators (in theory) but also deepens legal and contractual uncertainty, especially for those whose work is now more easily harvested, repurposed, and commercially exploited.

All attention now turns to the CJEU as we await what is set to be a highly anticipated and landmark ruling.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

The long-awaited opinion of Advocate-General (AG) Emiliou in C-590/23 Pelham, the enduring dispute between the electronic music group Kraftwerk against hip-hop producers, and their production company Pelham GmbH, has now been published. For those who had not been following the developments, the case has once again reached the Bundesgerichtshof (BGH), Germany’s Federal Court of Justice, which has submitted a new request for a preliminary ruling to the Court of Justice of the European Union (CJEU). The facts behind the dispute are by now familiar: in 1997, Pelham released the track Nur mir, which incorporated a two-second sample from Kraftwerk’s 1977 track Metall auf Metall, without the group’s consent. Given the strikingly unconventional tone of this opinion, this commentary will be divided into two parts. Part I sets the scene, offering context and examining the AG’s interpretation of pastiche. Part II will focus into the balancing of fundamental rights, highlight key gaps, and offer some concluding reflections.

In an earlier phase of the litigation, the CJEU held in Pelham (C-476/17) that the act of sampling (even very short audio fragments) falls within the exclusive reproduction right granted to phonogram producers under Article 2(c) of Directive 2001/29/EC (InfoSoc Directive). According to the Court’s reasoning, using such samples without prior authorisation can constitute an infringement of that exclusive right, unless the sample is unrecognisable in the alleged infringing use (which has given rise to a lot of commentary, see here, here and here).

The latest reference from the BGH asks whether unlicensed sampling might nonetheless be lawful when it qualifies as a “use for the purpose of pastiche” under Article 5(3)(k) of the InfoSoc Directive. The term “pastiche” is not defined in the Directive, and this referral gives the CJEU an opportunity to clarify whether certain forms of artistic reuse, such as music sampling, may fall within this exception. In essence, the case invites a re-examination of how copyright rules should apply to artistic practices that rely on transformation and reinterpretation rather than mere duplication.

The two questions referred to the CJEU are as follows:

• Whether the “pastiche” exception under Article 5(3)(k) of the InfoSoc Directive should be interpreted broadly to cover artistic reuse of existing works, such as sampling, without requiring specific elements like humour, stylistic imitation, or homage.
• Whether a use qualifies as pastiche only if intended by the user, or if it is sufficient that an informed audience recognises it as such.

Kiss your rights (as you knew them) goodbye…

AG Emiliou’s opinion stands out as a particularly engaging and far-reaching intervention in the ongoing conversation around copyright and artistic freedom. The opinion embraces an interdisciplinary lens, drawing not only on legal doctrine and scholarship but also on cultural theory and creative practice. It situates the legal question of pastiche within a broader cultural and technological context, one that spans both analogue and digital forms of expression and signals an awareness of the evolving dynamics of artistic production in the twenty-first century. In doing so, it goes beyond the specificities of the music recording industry to reflect more broadly on user-generated content, including references to memes and other digital forms of reinterpretation.

Before turning to the questions referred by the Bundesgerichtshof, the AG sets the stage with a reflection on the underlying tension in copyright law: while its purpose is to incentivise creativity, it may, under certain conditions, restrict it. This tension, AG Emiliou suggests, is particularly acute in contexts where creative reuse is central to artistic expression. In this light, the pastiche exception in Article 5(3)(k) of the InfoSoc Directive and rooted in the rights guaranteed under Articles 13 and 11 of the Charter of Fundamental Rights of the European Union (‘the Charter’) emerges as a potentially vital protection for the freedom of the arts.

According to AG Emiliou, there is increasing recognition that copyright law struggles to accommodate forms of creative reuse, particularly in artistic movements like appropriation and conceptual art. This tension is further exacerbated by the growing use of content moderation and automated infringement detection technologies, especially those introduced under Article 17 of the Directive on Copyright in the Digital Single Market (CDSM; for an overview of where we are 5 years later, see here), facilitating the control of short excerpts of copyright-protected works. For the AG, these mechanisms illustrate how copyright can impose a chilling effect on artistic expression, a problem compounded by the failure of existing exceptions to evolve in a way that adequately counterbalances these developments.

The framing here is significant. Rather than rooting the analysis in the language of exclusive rights (or the need for a high-level of protection for right-holders, seerecitals 4 and 9 of the InfoSoc Directive), the opinion emphasises the regulatory function of copyright within a democratic cultural space. One might note, with some curiosity, the absence of a property-based rationale in favour of a rights-balancing approach centred on artistic freedom until the near end of the opinion (article 17(2) of the Charter). This interpretive position suggests a shift in emphasis: not a mere balancing of artistic freedom and property rights on equal constitutional footing, but a view that recognises the evolving primacy of artistic freedom in light of how works are created, shared, and transformed in today’s digital landscape, and thus calls for a corresponding evolution in copyright law beyond internal mechanisms such as the idea/expression dichotomy.

Shine a Little Style…

Turning to the questions referred by the German court, AG  Emiliou begins by examining the scope of the pastiche exception under EU copyright law. As expected, he affirms that the concept of pastiche must be treated as an autonomous concept of EU law, requiring a uniform interpretation across all Member States. This interpretation should be guided by the ordinary meaning of the term, as well as the broader legal context and the objectives pursued by the exception itself. Furthermore, the opinion states that the exception does not necessarily have to be interpreted strictly, but rather that their effectiveness must be preserved (para 70). This is another indication of a shift in balance given that earlier case-law preferred to underline that strict interpretation did not mean restrictive interpretation and should ensure effectiveness of the provision (e.g. Painer, para 133; Deckmyn, para 22; ACI Adam and Others, para 23).

Rejecting the notion that the pastiche exception could serve as a general or open-ended clause (paras 71 and 75), an approach reminiscent of fair use doctrines, AG Emiliou cautions that such a reading could conflict with the three-step test outlined earlier in the opinion. Instead, he poses several essential characteristics that should define the scope and application of the pastiche exception.

In his interpretation of the pastiche exception, the AG draws on the CJEU’s judgment in Deckmyn (para 60), while carefully distinguishing pastiche from its related concepts of parody and caricature (para 62). Although the three are grouped together in the provision, the AG notes that pastiche is conceptually distinct, being primarily defined by the stylistic imitation of an existing work, genre, artist, or artistic school through the deliberate adoption of its characteristic aesthetic language. Notably, while clarifying this distinction, the AG also offers a more generous reading of parody than is sometimes assumed. This is illustrated in footnote 175, where he acknowledges that humour in musical parody need not always take the form of overt ridicule, but may instead emerge through more subtle or incongruous juxtapositions, for example, the unexpected combination of folk music with a death metal track. This suggests a broader interpretative latitude for both exceptions, supporting a more flexible understanding of creative reuse under EU copyright law. (I personally remain convinced that the CJEU has an opportunity to ensure that humour is interpreted in a broad sense to cover more creative reuses and respect a greater range of forms of humour, perhaps thereby obviating the need to distinguish pastiche from parody for copyright purposes, see here and here.)

Three core features of a pastiche are identified:

• first, it must recall a pre-existing work or artistic tradition by reproducing its stylistic markers;
• second, it must incorporate perceptible differences from the source, ensuring it is not merely a copy; and
• third, it must be intended to be recognisable as an imitation to those familiar with the referenced material.

Crucially, the purpose of the imitation, whether humorous, critical, or otherwise, is not decisive. What matters is that the new work clearly signals its referential nature and stylistic engagement with the original.

In one of the more critical clarifications of the opinion, AG Emiliou pushes back against a particularly expansive reading of the term “pastiche”. Although the German Government and the Commission had argued that pastiche should include any form of creative reuse, such as memes, mashups, or sampling, pointing to historical examples like the pasticcioopera of the eighteenth century (a genre made up of musical patchworks assembled from pre-existing arias) Emiliou draws a clear line. In his view, giving pastiche that kind of open-ended scope would overstate the weight of these alternative usages in ordinary language and, more importantly, overstate their influence on the drafters of the InfoSoc Directive. The implication is that not every act of reuse qualifies as pastiche simply because it borrows or assembles. What matters is whether the resulting work engages in overt recognisable act of stylistic imitation with perceptible differences from the original. As I’ve written elsewhere, a pastiche (like parody) is deeply connected to the original works reproduced and therefore, such uses require intentional reproduction of original works. Yet, the nature of the relationship is different as in a pastiche, there is less of detachment between the original and the use.

While the specific purpose behind a pastiche is not determinative (although it can be argued that the purpose of the copying is to create a pastiche), the user’s intent is essential, and that intent must be discernible in the resulting work. Pastiche is grounded in an intention to imitate or evoke the distinctive style of another work, artist, or genre. This intent must not be accidental or retrospective; it must guide the creation of the new work from the outset. The importance of intent ensures that the pastiche exception does not become a blanket licence for any form of reuse but rather supports artistic practices that consciously engage with and reframe existing works in a way that is identifiable and artistically meaningful (para 61).

Overall, despite good intentions, it is unfortunate that the purpose of creating is not more thoroughly examined. As the AG signals a shift away from the traditionally strong protection of exclusive rights, it becomes more important to ensure that this does not result in under-protection of copyright. Without a clear understanding of how uses like pastiche must meaningfully relate back to the original work, rather than treat it as a mere commodity, there is a risk that the balance intended by copyright law, between incentivising creation and enabling creative reuse, may be undermined.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

The role of copyright rules in the regulation of artificial intelligence (AI) continues to dominate legal discussions in 2025. With new case law, legislative initiatives, and regulatory developments unfolding across jurisdictions, the dialogue on how copyright interacts with AI models and systems – particularly around training data, authorship, and lawful use – has only grown more complex.

The recent preliminary reference to the Court of Justice of the European Union (CJEU) in Case C-250/25, Like Company v Google Ireland Limited from the Budapest Környéki Törvényszék (Hungary) illustrates this growing momentum (see early commentary by Keller, Guadamuz and Rosati). This reference comes on the heels of national first instance judgments on different aspects of text and data mining (TDM) and/or AI training in the Germany, The Netherlands, Hungary, as well as pending litigation in in France – by authors and publishers against Meta – and in Germany by collecting society GEMA against Suno AI and Open AI. Outside the EU, we can also mention the Getty Images v Stability AI case in the UK, a veritable tsunami of litigation in the US, as well a notable case  against OpenAI in India.

As we await guidance from the CJEU in C-250/25, it’s worth reflecting on the rich and diverse commentary published on the Kluwer Copyright Blog so far this year.

Below is a chronological round-up (in reverse order) of our posts from 2025 touching on AI, TDM, and related copyright issues:

June

Copyright, Compliance, and Confidentiality: Finding Common Ground in Generative AI —Enrico Bonadio, Eduardo Alonso and Vansh Tayal (Symbiosis Law School, Pune, India)
Read here

Report on LSE event – Artificial Intelligence, Intellectual Property, and the Creative Industries — Marianna Foerg
Read here

May

The EU’s false sense of isolationism in AI and copyright — Bertin Martens (Bruegel)
Read here

Do AI models dream of dolphins in lake Balaton? — Paul Keller
Read here

From Hollywood to Germany: A Transatlantic Comparison of Collective Agreements on AI in Film and TV Production — Christiane Stuetzle & Susan Bischoff
Part 1
Part 2

Does Human Learning Equal Machine Learning? High Court of Delhi to Rule on Lawfulness of TDM for Machine Learning— Pragya Jha & Bernd Justin Jütte
Read here

Third European Court Decision on the General Purpose TDM Exception Is Out — Peter Mezei
Read here

April

Originality in the Age of AI: How to Get Copyright Protection Now? — Jakub Wyczik
Read here

The TDM Opt-Out in the EU – Five Problems, One Solution — Martin Senftleben
Read here

March

Thaler v. Perlmutter: Human Authors at the Center of Copyright? — Matt Blaszczyk
Read here

Second and Third Drafts of the General-Purpose AI Code of Practice Released — Marianna Foerg
Read here

Remuneration for Use of Works in Text and Data Mining — Damian Flisak
Read here

Navigating User Lawfulness in European Copyright Law: From Lawful Use to Lawful Access — Tatiana Synodinou
Read here

Reconceptualizing the Reproduction Right in the Age of AI — Rita Matulionyte
Read here

Open Source Artificial Intelligence Definition 1.0 – A ‘Take It or Leave It’ Approach for Open Source AI Systems — Yaniv Benhamou & Michel Reymond
Read here

February

European Copyright Society Opinion on Copyright and Generative AI — Dusollier, Kretschmer, Margoni, Mezei, Quintais, Rognstad
Read here

The AI Act Provisions Relating to Copyright – Possibility of Private Enforcement? Germany as an Example— Jan Bernd Nordemann & Arman Rasouli
Part 1
Part 2

January

DPG Media et al. vs. HowardsHome – A National Ruling on DSM’s Press Publishers’ Rights and TDM Exceptions — Etienne Valk & Iris Toepoel

Read here

The Dubious Utilitarian Argument for Granting Copyright in AI-Generated Works — Patrick Goold
Read here

As the year progresses, we expect more key developments in this space, including judicial clarification, legislative action, and policy initiatives from global institutions. The Kluwer Copyright Blog will continue to monitor and contribute to the conversation.

Stay tuned for further updates!

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

Can European libraries lend digitised books? Comparative legal analysis

The fact that independent Secure Digital Lending (iSDL) may be allowed under EU law does not mean it is automatically permitted under the national laws of Member States. EU law enables this model but does not require countries to adopt it. While the Rental and Lending Directive, together with the VOB and Darmstadt rulings by the CJEU, open a legal pathway for digital lending by libraries, they do not obligate national legislatures to implement it.

For this reason, our study focused on a key question: To what extent do national legal systems across Europe currently allow—or could allow through dynamic interpretation—the implementation of iSDL by libraries? To answer this, we analysed the legislation of 22 countries, including 20 EU Member States, based on three essential legal conditions. These criteria reflect what is needed for iSDL to operate lawfully within national frameworks, considering EU directives and CJEU case law.

The first condition concerned whether national law provides a legal basis for digitising library collections (e.g., an explicit statutory provision authorising digitisation, a technology-neutral reproduction right that may cover digital copying, or an ancillary right). The second condition examined whether libraries are legally entitled to lend digital versions of works. While traditional (physical) book lending is universally recognised, e-lending remains legally unclear or restricted in many jurisdictions. National legal systems were assessed based on whether they expressly permit e-lending in legislation or can be interpreted, in light of the VOB ruling, to include digital formats within the lending right. The final condition concerned the existence of a Public Lending Right scheme that applies to e-books.

Findings: Fragmented national laws and the feasibility of iSDL

Although the iSDL model is legally allowed under EU law, our analysis shows no country has fully put it into practice. In none of the 22 studied countries do national laws meet all three conditions at once (Group 1). However, similar to R. Xalabarder’s research on Spain, the study found that in many countries, copyright laws can be interpreted more flexibly to allow e-lending. This goes beyond the exact wording of the laws and focuses on the purpose of copyright exceptions—to promote public access to knowledge and culture. The analysis divides countries into groups based on their copyright frameworks:

• Group 2 includes Germany, Croatia, Poland, and the United Kingdom, where national laws can be interpreted in ways that enable e-lending consistent with the VOB and Technische Universität Darmstadt
• Group 3, consisting of the Czech Republic, Estonia, Spain, Finland, France, Hungary, Ireland, Italy, Latvia, Lithuania, the Netherlands, and Slovenia, faces more interpretative challenges but these are not insurmountable.
• Finally, Group 4—Bulgaria, Greece, Portugal, Romania, Slovakia, and Ukraine—encounters more legal obstacles that currently prevent implementing iSDL e-lending.

Of course, interpretative approaches and doctrinal frameworks in individual countries mean that the classification we propose should not be considered definitive. However, it is important to note that, aside from Slovakia, local laws in most countries can likely be interpreted in a way that permits the implementation of e-lending under the iSDL model. This is because in Slovakia, book lending is not based on the derogation under Article 6 of the Rental and Lending Directive but rather on agreements with collective management organizations — making any lending model based on an exception effectively impossible.

Human rights as a foundation for Interpretation of copyright

One of the most important dimensions of the report is its emphasis on the need to interpret copyright law through the lens of users’ fundamental rights. The right to culture, freedom of expression, the right to education, and the right to privacy — all recognised under European and international human rights frameworks — must guide the evolution of copyright exceptions and limitations. In our view, implementing the iSDL model in European libraries requires a dynamic interpretation of copyright law—one that takes into account users’ fundamental rights. The CJEU, in the Funke Medien case, clearly emphasized that a national court must rely on an interpretation which, “whilst consistent with their wording and safeguarding their effectiveness, fully adheres to the fundamental rights enshrined in the Charter of Fundamental Rights of the European Union.” (para. 76)

The report highlights the key role of libraries and e-lending in supporting fundamental human rights like education, freedom of expression, cultural participation, and privacy. In today’s digital world, access to electronic books is crucial for equal opportunities and fighting digital exclusion. Libraries help people gain knowledge, develop skills, and connect with culture—important parts of the right to education and free information access. Using human rights to interpret copyright law creates a fair balance between protecting rights holders and meeting social needs, allowing libraries to carry out their mission online. Considering e-lending’s role in protecting fundamental rights when interpreting national copyright laws, following CJEU rulings, is not just optional but legally necessary. To comply with EU law, it may be needed to go beyond the literal wording of laws and rethink old concepts that have ignored this human rights dimension.

Digital publishing and user privacy: A conflict of models

A key factor supporting the legality of e-lending under the iSDL model is privacy. More and more, libraries provide e-book access through platforms owned by publishers or commercial companies. While this allows fast and wide access to digital content, it raises serious privacy concerns and may conflict with public libraries’ core values. Using e-books in this way involves collecting and sharing user data. Problems occur when license agreements force libraries to share this data—usually with rights holders. Even if this complies with data protection laws, it involves not just personal data under GDPR but also other data about how e-books are used.

Many publishers, including academic ones, are shifting to data-driven business models where user data is as valuable as the content itself. Apps and websites for e-books collect information like reading time, highlights, search terms, and user habits. This clashes with librarianship’s ethical principles, which focus on protecting user privacy and allowing access without surveillance. This “surveillance publishing” model tracks user behavior and threatens a core value: intellectual freedom—the right to read, research, and explore information anonymously.

Users should be able to use library resources without being tracked and decide if their data stays within the library or is shared. Libraries should not be forced to act as middlemen in commercial data collection. The iSDL model can solve this by ensuring all interactions and data stay only between the library and its users, without involving third parties. This protects privacy, keeps access under library control, and supports intellectual freedom.

Policy recommendations: Towards a sustainable e-lending

Based on the study’s findings, we have developed recommendations across three key levels. To ensure uniform e-lending access across Europe, the EU legislators should introduce a mandatory copyright exception allowing libraries to offer e-lending under the iSDL model or other approaches consistent with the VOB ruling. This exception must be technology-neutral and not override contracts or TPMs. It should also permit libraries to digitize and use e-books even if current agreements or technologies restrict this. Authors should receive fair remuneration through the Public Lending Right scheme. Ideally, this exception would be part of a broader framework focused on access to knowledge, such as a Digital Knowledge Act or European Research Area Act.

Regardless of EU legislation, national lawmakers should, within EU limits, introduce provisions allowing iSDL in their legal systems. Such regulations are essential to protect users’ fundamental rights, currently weakened by legal uncertainty and exclusive reliance on commercial licenses. Libraries need legal certainty to engage in e-lending regularly, best achieved through clear legal frameworks.

Finally, achieving e-lending requires the active involvement of libraries themselves. Whenever possible, they should implement e-lending within the framework of existing national laws. Local library associations must develop common guidelines addressing the legal and technical aspects of e-lending. While many national laws can be interpreted in ways compatible with iSDL, adopting codes of practice would provide greater clarity and legal certainty. Additionally, libraries need to actively engage with policymakers, demonstrating how current market conditions undermine their mission and the rights of users.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

In an era where digital access to knowledge shapes the frontiers of education, research, and participation, European libraries face significant legal and technical obstacles in lending electronic books. While the shift from paper to digital is well underway in many sectors, libraries — long seen as guardians of knowledge and enablers of equal access — are navigating a legal labyrinth when it comes to e-lending. A recent report led by the Future Law Lab at Jagiellonian University and the Centrum Cyfrowe Foundation, developed as part of the KR21 project, highlights this pressing issue and proposes a pragmatic, legally sound solution: the independent Secure Digital Lending (iSDL) model.

At the heart of this model is a legal and ethical argument: libraries must not be sidelined in the digital transformation due to outdated or overly restrictive interpretations of copyright law. Instead, they should be empowered to digitise legally acquired print books and lend them digitally within a secure framework. This approach preserves the continuity of their public mission in the digital age while safeguarding users’ fundamental rights to culture, education, information, and privacy.

The report was developed in response to two key questions:

1. Is e-lending under the iSDL model permissible under international and EU law? (Part I)
2. Do the national laws of selected European countries allow libraries to implement iSDL? (Part II)

Two legal paths to e-lending: licences vs. copyright flexibilities

Digital lending in libraries follows two main legal paths. The first is based on licensing agreements with publishers. In this model, libraries access e-books through contracts that define strict conditions—such as loan duration, number of users, and technical restrictions. While legally straightforward, this model limits library autonomy. It makes them dependent on commercial terms, which may restrict access, compromise user privacy, and undermine long-term access to digital collections.

The second path relies on copyright flexibilities—exceptions and limitations that allow certain uses without prior permission. From this perspective, public and academic libraries should be able to digitise books they lawfully own and lend them electronically, as long as they follow specific legal conditions. These include lawful source, time-limited access, and compliance with the three-step test, which balances user rights with the legitimate interests of authors.

CDL vs. iSDL: Two models of e-lending based on copyright flexibilities

Two distinct models of e-lending based on copyright exceptions have emerged: Controlled Digital Lending (CDL) in the USA and iSDL in Europe. Though technologically similar, they differ significantly in legal basis and justification. Both models allow libraries to digitize legally acquired physical books and lend digital versions using a one-copy-per-user system, with time-limited access and technical safeguards. CDL has been challenged in U.S. courts, notably in a case brought by publishers against the Internet Archive. While both CDL and iSDL use similar lending methods, their legal foundations are different. CDL relies on the U.S. First-Sale Doctrine and Fair Use, and does not require additional remuneration, assuming the purchase price covers lending. By contrast, iSDL is grounded in EU law—specifically Article 6 of the Rental and Lending Directive and Article 5(2)(c) of the InfoSoc Directive. iSDL also includes the Public Lending Right (PLR). These differences mean that legal arguments used against CDL in the Internet Archive case cannot be directly applied to iSDL.

Understanding e-lending through international law

When analyzing e-lending under international law, the WIPO Copyright Treaty (WCT) is especially relevant. Although the treaty doesn’t directly regulate lending rights, it offers key guidance for understanding where e-lending fits in the broader legal context.

Article 7 of the WCT addresses the Right of Rental, but it applies only to specific types of works and physical copies—not to digital lending. Article 6, which governs the Right of Distribution, allows countries to define the conditions under which this right is exhausted. However, the most important article for digital lending is Article 8: the Right of Communication to the Public. This gives authors exclusive control over how their works are made accessible, including on-demand access—an essential element of e-lending.

In short, physical lending relates to Article 6, where rights may be exhausted and supported by exceptions under the three-step test in Article 10. Digital lending, however, is covered by Article 8, where the exhaustion principle does not apply. Still, e-lending may be allowed under specific exceptions and limitations—if they meet the conditions of the three-step test. The report provides a detailed assessment of how the iSDL model aligns with this test.

The qualification of e-lending from the perspective of EU law

European copyright law does not explicitly regulate e-lending. Before the VOB ruling, it was generally seen as possible only through licensing agreements. This was because EU law lacked specific exceptions for e-lending. The practice was treated as part of the “making available to the public” right under Article 3 of the InfoSoc Directive. Additionally, the term “copy” was traditionally interpreted to mean only physical, tangible objects.

The landmark VOB judgment significantly changed how lending rights are interpreted. It adopted a broader view, stating that there’s no clear reason to exclude digital copies from the scope of the Rental and Lending Directive . This means a “copy” can also be digital. This distinction matters because lending rights are based on different legal grounds than rental rights and allow more flexibility.

Crucially, if the Court had classified e-lending as “making available”, libraries would need licenses from rights holders, since no exception exists for that under the InfoSoc Directive. Instead, the ruling made it possible to allow e-lending under Article 6 of the Rental and Lending Directive, which significantly reshapes the legal basis for digital lending in libraries. According to the ruling, the concept of lending:

covers the lending of a digital copy of a book, where that lending is carried out by placing that copy on the server of a public library and allowing a user to reproduce that copy by downloading it onto his own computer, bearing in mind that only one copy may be downloaded during the lending period and that, after that period has expired, the downloaded copy can no longer be used by that user (para. 54).

VOB ruling: open issues and practical problems

Although this ruling was a major breakthrough in recognising that digital lending by libraries can fall under EU lending rights, it left several important questions and practical challenges unanswered. One of the key issues is how libraries can legally obtain digital copies of books for e-lending. There are two main approaches, each with its own complications.

1. Licensing from publishers

In theory, libraries can legally acquire e-books by purchasing them from authorized sources. However, this becomes problematic when the accompanying licenses prohibit e-lending or when the files are locked with technical protection measures (TPMs). The VOB ruling does not explicitly grant libraries the right to override license terms or circumvent TPMs. Moreover, in practice, even acquiring e-books can pose significant challenges. Publishers may outright refuse to license digital books to libraries or impose restrictive conditions—such as charging prices far higher than those available to individual consumers. Frequently, libraries are prevented from purchasing individual titles and are instead offered large, bundled collections that may not align with local needs. These limitations severely restrict libraries’ autonomy, making it difficult to curate collections based on user demand and fulfil their public mission in the digital environment.

However, it is also conceivable that publishers might offer individual e-books to libraries at reasonable prices, with licensing terms that do not restrict e-lending and without applying TPMs that would block it. For this reason, this model of e-lending is best described as dependent Secure Digital Lending—it relies heavily on publishers’ goodwill to provide access on fair terms. Without such cooperation, libraries face significant legal and practical obstacles in delivering effective e-lending services.

2. Digitisation of print books

Another possible route for libraries is to digitize the paper copies they already own. However, this raises a distinct legal question: Is digitization for the purpose of e-lending permitted under EU law? The VOB ruling did not provide a clear answer, leaving some legal uncertainty around this practice.

In the report, we present an argument supporting the permissibility of libraries digitizing books to conduct independent Secure Digital Lending. Our approach draws inspiration from the Opinion of Advocate General Szpunar in the VOB case, who stated that:

the exception under Article 5(2)(c) of the same directive ought to come into play to enable libraries to benefit from the derogation from the lending right provided for in Article 6(1) of Directive 2006/115 (point 57)

This aligns with the Court’s earlier ruling in the Technische Universität Darmstadt case. There, the CJEU determined that the exception in Article 5(2)(c) can be used to enable libraries to make works available to the public under Article 5(3)(n). The report analyses the extent to which applying similar reasoning to e-lending effectively establishes a legal framework that permits libraries to create the digital copies necessary to carry out the e-lending process under copyright exceptions. This approach forms the basis for recognizing independent Secure Digital Lending or iSDL as lawful under EU law.

The second part of this post will explore whether the national laws of selected European countries allow libraries to implement iSDL.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

The rise of generative AI and automated content generation has raised legal and ethical issues, making them a focal point in creative and technological sectors. As stakeholders navigate this new terrain, the EU AI Act appears as a benchmark regulatory framework. This blog briefly examines the transparency provisions and trade secret protection under the Act, not as sterile concepts of law but as relevant issues that affect interactions between creators, developers, and regulators in generative AI.

By framing the discussion through such practical lenses as artistic integrity, market competition, and societal trust, actionable insights are provided for legal practitioners and technologists. We here concisely highlight the opportunities and challenges these provisions present, especially vis-à-vis the balancing of copyright protections and innovation. Ultimately, the piece intends to help rights holders and AI developers proactively confront potential legal friction by encouraging meaningful discourse and steering the future of generative AI to honour the twin pillars of creative expression and technological progress.

The Opacity Challenge and the EU AI Act: Transparency vs. Trade Secrets

It is commonly noted that generative AI models typically act as a mysterious black box. The catch is that a person can observe the inputs and outputs, but not the main reasoning behind them. This opacity makes it strenuous for copyright owners to know if, when or how their works were ingested or somewhat used without permission for AI training purposes.

The EU AI Act has taken a pioneering step by requiring providers of general-purpose AI models to provide sufficiently detailed summaries regarding the data upon which their training was carried out. At the same time, AI companies worry that such transparency obligations would be expensive and expose proprietary architectures and trade secrets, jeopardising innovation and competitiveness.

The relevant provision of the EU AI Act is Article 53(1)(d), requiring providers of general-purpose AI to prepare and publish a ‘sufficiently detailed summary’ of the dataset used for training, according to a template established by the AI Office. The dataset summary, provided in Annex XII, must contain at least information on content categories, sources, volume estimation, and methods of processing involved, but does not entail sharing raw data files.

Copyright holders view these summaries as a pivotal tool for discovering unauthorised usages and requesting compensation. Yet, as mentioned, AI developers argue that a wide blanket disclosure would disclose crucial aspects of model architecture, training methodology and proprietary pre-processing pipelines; and small and medium-sized enterprises fear that the cost and complexity of preparing these summaries might even prevent their capacity for innovation. Although the EU AI Act provides some SME-friendly measures facilitating access, priority fee-exempt participation to regulatory sandbox, simplified documentation templates, and proportionate assessment fees, this strict set of rules might entail increased costs, stronger bureaucracy, and hurdles at market entry, and consequently limit the involvement of smaller players in the AI ecosystem.

To alleviate these pressures, Recital 107 of the EU AI Act allows companies to benefit from a trade-secret defence for withholding pipeline or algorithmic details provided that the summary is generally comprehensive in its scope instead of technically detailed. The trade-secret exemption has been claimed by industry groups to be paramount in protecting innovation, posturing that divulging too much detail will hand competitors ‘the keys to the kingdom.’

In contrast, full-transparency advocates caution that in the absence of well-defined, case-by-case limitations and independent review, the trade-secret defence could devolve into a ‘blanket excuse’ for intransparency, thus undermining the very goals of accountability enshrined in the Act. Indeed, several vendors of AI-based products exploit the narrative of ‘inherently unexplainable AI’ to justify very few or minimal disclosures. In response to this self-serving stance, the European Data Protection Board has asserted that black-box opacity cannot constitute a valid ground for avoiding transparency obligations either under the GDPR or under the EU AI Act. This could imply that certain platforms may be taking unscrupulous advantage of the claim of inscrutability to exercise delay or to circumvent their reporting obligations. This is also why, it has been argued, secrecy claims must be strictly circumscribed and justified by way of a public interest test.

Practical Compliance Strategies

Since this tension exists, it is important to come up with templates for transparency that protect real trade secrets while furthering the EU AI Act’s intention to empower rights-holders and build trust. Given this need to reach a balance between transparency and confidentiality, what practical steps can platforms take to meet the Act’s mandates? First, in accordance with the EU AI Act, Article 53(1)(d), these entities may produce high-level summaries of training data that mention general classes of sources without revealing proprietary details. According to Recital 107, dataset providers may employ broad-type descriptions of the datasets; for instance, web scraping social media posts or licensing books and articles to avoid specifying each URL, individual title or database entry.

The exclusion of proprietary model details is an intentional feature of the EU template. It requests information on data sources and processing but does not request algorithms and architectures or exact pre-processing methods. For example, a company might record, ‘The model was trained on ~10B tokens from public news sites (2015–2023), Wikipedia dumps, and a cleaned subset of social media text.’ Such a statement would be sufficiently detailed and thus would enable rights holders to understand the broad scope of the content while protecting the company’s investments in the content. Put another way, emphasising just the source types and the general methods used in preparation affords perfect compliance without giving away trade secrets.

For example, the Foundation Model Transparency Index developed by researchers at Stanford University evaluates whether a major model really fulfils the criterion: Meta’s Llama 2 scores just 54 percent, and OpenAI’s GPT-4, a mere 48 percent, on 100 transparency metrics. The very low grades can somehow indicate the evident gap that exists between legal obligations and actual behaviour, advocating for more viable measures of transparency.

Beyond dataset summaries, another approach could be to adopt standardised model cards as a pragmatic middle ground. First proposed in 2018 by Mitchell et al., model cards accompany trained AI models with documentation that explains their intended use cases, categories of training corpus, metrics used in measuring performance, and any known shortcomings. The cards promote transparency by divulging high-level information regarding data sources, without the card issuer revealing any details deemed proprietary. Organisations like Google have developed model cards on their AI platforms to help end users assess model suitability. Privacy and governance bodies such as IAPP recommend model cards to promote responsible AI deployment. AI providers may want to extend this approach with a dedicated copyright-transparency section, broadly listing data sources, such as license-holder publisher archives, public domain texts, or user-uploaded content, without giving exact file paths or storage locations. Such customised model cards would enable rights-holders to self-identify potential uses of their works, balancing the need for oversight and trade-secret protection.

A partnership between rights holders and the AI developers could also be considered to build technical safeguards. An idea would be a public registry of all known copyrighted content or watermarked models. For instance, policy experts propose ‘standardising watermarking and maintaining a registry of watermarked models and detection services’ whereby users can easily check any content. By this analogy, publishers would be able to register digital fingerprints (hashes) of their works in a common database. AI labs may then use automatic matchers, for example, perceptual hashing or machine-learning classifiers, to flag protected inputs in the training data or outputs against that registry.

Alternatively, consortia of creators may want to get select access to model Application Programming Interfaces (APIs) with copyright-compliance tests or work on digital rights registries, like those used for music. For instance, US collecting society SoundExchange is formulating a worldwide AI registry that would allow rights holders to opt in or out of having their sound recordings used for training. Such collaborative infrastructure could facilitate identifying copyrighted material used in training or created by an AI system, thereby providing rights holders with a method to monitor and enforce their rights besides merely concerning themselves with data collection.

Conclusion

We have seen that the tension between transparency obligations in the EU AI Act and protection of trade secrets represents a challenge. Finding the right balance requires a balanced approach that satisfy both copyright holders’ legitimate interests and developers’ innovation concerns. High-level dataset summaries, standardised model cards with copyright sections, and collaborative registries may offer practical compromises that honor the spirit of regulatory accountability without exposing proprietary technologies. As implementation proceeds, regulators must ensure trade secret exemptions remain narrowly defined while industry establishes best practices for meaningful disclosure. Only through a measured approach can the AI ecosystem foster both innovation and trust, creating sustainable paths forward for all stakeholders.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

On 4 March 2025, the London School of Economics (LSE) hosted an event dedicated to exploring the intersection of AI and IP within the creative industries. The discussion has been prompted by the recent developments in this area in the EU, the UK and the US. Four distinguished academics shared their insights on this complex and evolving topic.

The Generative AI Question

Martin Kretschmer opened the discussion by highlighting the main trends in generative AI. He emphasised that reproductions occur at various stages of the AI lifecycle, from data scraping to output generation, prompting legislators to steer the development of technology to address these challenges.

The UK aims to align its system with the EU by linking obligations at the data scraping and collecting stage (the AI Act) with exceptions from the Copyright in the Digital Single Market Directive (CDSM Directive). This involves providing sufficiently detailed summaries of the training material for AI models and clarification of the opt-out mechanisms, ensuring training activities are only possible if rights have been reserved (Article 53(1) and Recital 105 of the AI Act).

Reproductions During Training and Output Stages

The question of whether reproductions of copyright works occur at the training stage and, if so, whether this constitutes copyright infringement, is a significant point of debate. One needs to consider whether the outputs generated by AI are substantially similar to copyright works and whether such outputs are potentially infringing.

At the end of February 2025, the UK government closed a consultation on copyright and AI, proposing a data mining exception (see Option 3 below) that allows rights holders to reserve their rights, supported by transparency measures. This approach is similar to the EU’s framework (the AI Act).

Licensing Agreements and Legal Trends

Regardless of the legal system, there is a rise in licensing deals, which can be linked to the adoption of the AI Act in the EU and a sharp increase in litigation in the US. Interestingly, Professor Kretschmer pointed out that “new” content creation and distribution companies make the majority of licensing deals, with the news and media sector dominating.

Economic Impact on Creatives

There is a general trend towards licensing due to legal uncertainty and a drive towards a licensing economy. This trend has contributed to the halving of income for artists and literary authors, even before the advent of AI. Factors such as the platform economy and oversupply of services exacerbate this issue. With AI, this trend is expected to continue, and professional career options for creatives remain bleak. The current opt-in/out mechanisms are unlikely to improve the situation, necessitating alternative solutions to enhance the position of creatives.

Copyright and AI Consultation in the UK

Tanya Aplin focused on the UK government’s consultation on copyright and AI, which closed on 25 February 2025. She outlined the options proposed by the UK Intellectual Property Office (UKIPO):

• Option 0: Do nothing. The UKIPO deemed this option undesirable, not least because something needs to be done in respect of the current text and data mining (TDM) exception in Section 29A of the Copyright, Designs and Patents Act (CDPA).
• Option 1: Licensing in all cases, leaving it entirely to the market without exceptions. The UKIPO thought this option was also unwise.
• Option 2: Broad data mining exception, allowing TDM for any purpose with lawful access without rights holders’ permission. Professor Aplin thought that following the Hargreaves report, there is generally an opportunity for introducing a fair use exception, but this needs a wider consultation regarding the structure and scope for the UK exceptions with the wider scope of stakeholders and users.
• Option 3: Introduce a TDM exception with lawful access, underpinned by transparency. This would permit TDM for any use by anyone, but rights holders would be able to opt-out individual works, sets of works or all of their works that they do not want to be used for commercial purposes. This is the UKIPO’s preferred option, as it provides a balance between encouragement of AI development and copyright. However, the opt-out mechanism has a level of complexity, unworkability and nuance. Relying on opt-outs may also conflict with the international copyright law.

Professor Aplin also discussed some other options reflected in literature.

Compulsory Licensing (Christophe Geiger and Vincenzo Iaia, see here: Part 1 and Part 2)

This approach would involve introducing a statutory remuneration right for the use of works protected by copyright for commercial machine learning purposes in place of the opt-out right under Article 4(3) of the CDSM Directive, ensuring that issues are handled in a manner compliant with fundamental rights. However, this solution faces significant challenges, particularly in calculating the value of the works used. Therefore, this approach may be too extreme and difficult to implement effectively.

Levy System (Martin Senftleben, see here)

Another option is to move away from statutory licensing and adopt a levy system applied to AI tools themselves. Levies could be applied uniformly, either as a lump sum or a percentage of revenue, with licensing organisations responsible for collecting these levies. This system aims to address issues of trust and transparency. However, there are concerns that the levy system may not work effectively, as evidenced by its complications in the context of private copying. Implementing such a system for AI tools could prove even more challenging.

Amending Section 29A CDPA

This section currently restricts TDM to non-commercial research and does not allow for the transfer of copies or knowledge exchange. It also does not apply to database rights. Professor Aplin’s proposal is to amend Section 29A to widen its scope to include any type of scientific research and extend its application to database rights. This option, preferred by Professor Aplin, aims to enhance scientific research using AI while addressing legal uncertainties and potential workarounds. However, she thought there was legal uncertainty regarding what would constitute scientific research purposes, which still needs discussing.

Disruption Caused by Generative AI

Luke McDonagh addressed the disruption caused by generative AI, particularly in the creative industries.

The Impact on Creative Industries

Generative AI has led to a decline in employment opportunities for voiceover actors, as AI technologies can replicate and alter voices.

For instance, OpenAI allegedly used Scarlett Johansson’s voice without her permission, leading to a dispute, which then settled. Similarly, David Attenborough’s voice has been cloned. These examples highlight the extent to which AI can modify creative outputs post-production.

Legal Protections and Challenges

In the UK, there is no specific personality right, resulting in a patchwork of protections for performers’ rights. The law relating to performers’ rights is limited in scope, making it difficult to enforce rights over one’s voice or image. While there is potential for the UKGDPR protection (which replaced the GDPR following Brexit), its scope is also limited. The most promising legal avenue for people in the UK to protect their own image appears to be the law of passing off. However, passing off requires meeting the goodwill standard and proving misrepresentation and damage, which can be challenging.

The Need for New Statutory Rights

Given the limitations of existing legal frameworks, there is a growing discussion about whether new statutory rights are needed to address the gaps in protection. Established practices for digital replicas currently exist, but the scope of these rights requires careful consideration.

US Publicity Rights and AI

Madhavi Sunder discussed the US perspective on publicity rights in the context of AI.

Right of Publicity in the US

When OpenAI asked Scarlett Johansson to be the voice of ChatGPT’s “Sky” and she refused (see, e.g. The New York Times), they used a similar voice anyway, possibly violating her right of publicity. In the US, the right of publicity is a tort recognised in most states, allowing individuals (mainly celebrities) to control the commercial value of their identity. Right of publicity is a broad legal concept that includes name, image, and likeness. There is no single, federal right of publicity, but instead a collection of state-level rights that together protect an individual’s identity. Right of publicity is particularly strong in California. California Assembly Bill 1836 updated Section 3344.1 of the California Civil Code, which governs the post-mortem publicity rights of deceased individuals.

The Need for Stronger Protections

The Johansson case underscores the need for stronger protections against digital replicas, such as deepfake pornography. The Take it Down Act initiated by Melania Trump is aimed at speeding up the removal of non-consensual intimate imagery (NCII) and “deepfakes” online. Although drafted with good intentions in mind, the Act could lead to overreach and censorship.

Legislative Efforts and Proposals

New federal legislation is needed to address the challenges posed by digital replicas. Two notable pieces of proposed federal legislation aim to strengthen protections:

1. No AI Fraud Act: This Act would create a federal right in voice and likeness, protecting against deepfakes and allowing rights to be transferred during an individual’s lifetime. It includes secondary liability for social media hosts and balances free speech considerations.
2. No Fakes Act: This Act would establish a federal right to control image, voice, and likeness, applying to all individuals, not just celebrities. It includes freedom of expression exclusions and requires platform providers to remove materials once notified.

There is also a push to extend copyright protections to traditionally non-IP matters, such as deepfake pornography.

Conclusion

The event at LSE underscored the need for thoughtful and balanced approaches to AI and IP within the creative industries. As AI continues to evolve, policymakers must navigate the complexities of copyright, performers’ rights, and publicity rights to ensure fair and equitable outcomes for all stakeholders. A recording of the LSE’s event is available to view here.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

On 9 April 2025, the Supreme Constitutional Court of Cyprus handed down an Opinion concerning the constitutionality of two amendments to Section 26 of Law 65(I)/2017 on collective rights management and the granting of multi-territorial licenses for online use of musical works (Reference No. 5/2024). The amendments had been enacted by the House of Representatives in July 2024, but were remitted by the President of the Republic in August 2024. The issue was then referred to the Supreme Constitutional Court for its opinion.

Background

A significant contribution of the acquis communautaire in terms of Cypriot copyright law is the harmonisation of rules governing collective management, as effected by Directive 2014/26 on collective management of copyright and related rights. This was given effect under Cypriot law by Law 65(I)/2017.

The Law 65(I)/2017 has been fraught with challenges from the start. It was hastily enacted in order to comply with the deadline for transposing Directive 2014/26/EU and made reference to Regulations, which were meant to provide more practical detail. For example, Regulations are needed for fixing the application fee for CMO registration (s. 13(1)(b)) and the renewal fee (s. 51(2)(c)), determining the hearing procedure for complaints against collective management organisations (CMOs) (s. 43(5)), enacting a procedure for the operation of the supervising authority (s. 51(2)(a)), and determining the content and ways to publicise the Register(s) of CMOs (s. 51(2)(b)). The speed with which the primary legislation was passed meant that these Regulations were not enacted at the time. This meant that key provisions of the legislation were rendered very difficult – or even impossible – to enforce; indeed, even today, basic matters such as the registration process for CMOs are currently in limbo.

Negotiations have been underway on this topic in the Committee of Commerce and Industry of the Parliament of the Republic of Cyprus for years and these Regulations have still not been issued. However, on 6 April 2023, an amendment to the Law was tabled with the Parliamentary Committee of Energy, Commerce, Industry and Tourism. On 11 July 2024, the amendment was adopted by the plenary session of the House of Representatives, but it was remitted by the President of the Republic of Cyprus in August 2024. The matter was referred to the Supreme Constitutional Court by the President of the Republic for its opinion as to whether the two legislative amendments made to section 26 of the Law 65(I)/2017 were unconstitutional.

The amendments in question

Article 16 of Directive 2014/26/EU relates to the licensing mechanisms available to CMOs and provides that licensing terms must be based on objective and non-discriminatory criteria. This was transposed into Cypriot law by Section 26 of the Law. The amendments in question related to this section and concerned the following:

1. An amendment that would allow users who dispute the CMO’s set fees to pay what they consider to be a reasonable fee and deposit half of the disputed amount into a special fund, where it would be frozen until the resolution of the dispute.
2. An amendment that would allow users to pay a lump sum annually (or as otherwise agreed) to a single CMO, as the fee corresponding to each right for which it has been licensed.

The position of the President of the Republic of Cyprus – who had referred the matter to the Supreme Constitutional Court – was that the amendments in question were contrary to Articles 1Α (supremacy of EU law over national law), 23 (protection of the right of property), 26 (protection of the freedom to contract), 28 (equality of all persons before the law) and 179 (supremacy of the Constitution as a source of law with the exception of Article 1Α) of the Constitution.

The Court’s analysis

The first amendment in question

The Court began by pointing out that there is no distinction between absolute and exclusive rights on the one hand, and the rights to equitable remuneration on the other. It went on to say that in the event of a dispute, it is the users who determine the remuneration; furthermore, regardless of the nature of the right, users are indiscriminately exempted from the obligation to pay the remuneration, which the rightsholder is (provisionally) demanding.

At the same time, the Court opined that CMOs, as authorised managers of the rightsholders’ rights, are deprived of the right to issue a reasoned refusal, which serves as a counterbalance to the principle of the organisation’s obligation to enter into a contract. This contradicts section 26(3)(b) of the Law 65(I)/2017 (Article 16(3) of the Directive), which provides that CMOs shall reply without undue delay to requests from users, indicating, inter alia, the information needed in order for the CMO to offer a licence.

The Court then found that the first amendment in question is a restriction on intellectual property rights protected by Article 17(2) of the Charter of Fundamental Rights of the European Union and movable property rights protected by Article 23(2) of the Cypriot Constitution. Moreover, no persuasive ground for such a restriction had been put forward by the House of Representatives; on the contrary, it disrupted any existing balance within the Law in favour of the users.

It was, therefore, held that the first amendment in question was contrary to Article 23 of the Constitution.

The second amendment in question

The Court set out the framework for the operation of CMOs and went on to describe the essence of the second amendment as an attempt to allow any CMO, at the discretion of any third party or user, to collect and distribute sums on behalf of any other third party CMO – once again without any explanation put forward.

The applicant’s position was that this was contrary to the constitutional right to freedom of contract. The position of the House of Representatives was that the user’s obligation to pay the total amount of remuneration to all CMOs (e.g., CMOs representing composers, CMOs representing performers, etc) remains, and that it is only for the sake of convenience that this amount will be paid to a single CMO, which then has the obligation to distribute the amounts to the other CMOs. This, counsel for the respondent claimed, is intended to facilitate the overall system, without affecting the rightsholder’s core rights; after all, the Law already recognises the transfer of amounts from one CMO to another in relation to deductions and payments provided for in representation agreements (section 25 of the Law; Article 15 of the Directive).

The Court found that the assignment of the collection and distribution of the remuneration falls within the scope of the authorisation or contract of assignment between the rightsholder and the CMO, which the rightsholder has the absolute right to choose. The freedom to contract includes the freedom to choose one’s counterparty.

With the second amendment in question, rightsholders are required to accept the payment of their remuneration through a third-party CMO and not through the organisation they themselves selected and established a contractual relationship with, expecting to receive their rights from it. This third-party CMO would have no contractual relationship with such rightsholders. Nor would it bear any of the obligations imposed by law on the organisation authorised by the rightsholder to represent their interests, such as obligations of trust, accountability, and transparency. This leads to the blatant violation of the contractual relationship between the rightsholder and the organisation that the rightsholder had the absolute contractual right to choose. The Court found that this constitutes a breach of the rightsholders’ right to freedom of contract.

Moreover, the Court went on to say, the third-party CMO is burdened with an obligation that does not arise from its contractual relationships with the rightsholders that chose it and whose authorisation it has accepted. Rather, this obligation arises from the contractual obligations of another organisation (i.e. the one that the beneficiary originally selected). The Court found that the third-party CMO’s constitutional right to freedom to contract is also violated.

Consequently, the Court held, the second amendment in question is contrary to Article 26 of the Constitution.

Commentary

This is an important case for many reasons. First, it is the first time the Supreme Constitutional Court of Cyprus has dealt with the Law 65(I)/2017. Given the scarcity of the case-law related to copyright and related rights and the nascent status of collective management in Cyprus, this is a milestone.  The Opinion clarifies that rightsholders’ rights must be balanced with users’ rights; it also affirms that copyright and related rights constitute property under the Constitution (Article 23) and EU law (Article 17(2) of the Charter of Fundamental Rights). As the Court stressed, copyright and related rights should, according to recitals 11 and 12 and the whole spirit of the Directive 2001/29 benefit from a high level of protection.

It is also noteworthy that the Court places a strong emphasis on the protection of contractual freedom by the Constitution and more specifically on the freedom to choose the contracting party. In Cypriot constitutional law, the protection of contractual freedom is not considered to be absolute and, specifically, three exceptions are set: (i) the general principles of contract law; (ii) the public interest; and (iii) the Constitution itself. The absolute manner in which this freedom is protected in this Opinion might raise some doubts on the compatibility of mechanisms of extended collective licensing and mandatory collective rights management with Article 26 of the Constitution. With respect to collective licensing with extended effect, Article 12 of  Directive 790/2019 has been transposed almost verbatim in section 33 of Law 59/1976. However, since the Constitution of Cyprus established the supremacy of EU law even over the Constitution itself (Article 1 A of the Constitution), Article 12 of Directive 790/2019 may take precedence over Article 26 of the Constitution. Nonetheless, the same reasoning is not possible for any other potential mechanism of collective licensing with extended effect in the future, which would have resulted from the initiative of the Cypriot legislator.

Furthermore and from a more practical perspective, it is also curious that the legislator saw a need to improve the enforcement of the Law on Copyright collective management when, in fact, the absence of the Regulations for the past eight years has meant that the most effective dispute resolution mechanism – that of the so-called Competent Authority, established per the requirement under Article 36 of Directive 2014/26/EU – was inactive and so it was not really possible to assess whether a need for improvement existed. Arguably, the legislature’s priority ought to have been pushing through the Regulations rather than tweaking the Law before it had a chance to be properly enforced.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

After having introduced the issue and restated the legal framework and the facts of the case, the AG addressed each issue raised by the referring courts in turn namely the relationship between copyright and design protection, the criteria for assessing originality and the criteria for assessing copyright infringement.

Relationship between copyright and design protection

It is good that the AG spends some time (almost as much as on infringement) on this issue. The main question was whether higher requirements of copyright protectability must apply for works of applied art. In Cofemel, the Court replied in the negative but one of its statements (paragraph 52: “although the protection of designs and the protection associated with copyright may, under EU law, be granted cumulatively to the same subject matter, that concurrent protection can be envisaged only in certain situations”) left a doubt. After having recalled the Court’s case law on originality, the AGconcludes that the statement at paragraph 52 of Cofemel is “a simple reminder to the national courts that there is no automatic connection between the grant of protection under design law and protection under copyright law and that the conditions for such protection, namely novelty and individual character on the one hand, and originality on the other, must not be confused” (para 36). In addition, the said paragraph does not mean that a higher requirement of originality must be applied to works of applied art to limit their protection by copyright, because it would also run against the Court’s statement in paragraph 48 of Cofemel and the general scheme of the judgment that the same criterion of originality applies to all works.

Originality

The referring courts had asked what criteria should be taken into account to determine originality.

Firstly, the AG states that the application of the criterion of originality must take into account the specific nature of the type of works concerned. For him, compared to other works, there is no presumption of creativity for utilitarian works (para. 42). Another important clarification is terminological. He rightly says that confusion can ensue when a court uses the terms ‘aesthetic’ or ‘artistic’ in relation to the choices made by the author. In some senses those two terms can imply that the work is creative, but not always. He thus recommends using the terms ‘free and creative choices reflecting the personality of the author’ and not the terms ‘artistic’ and ‘aesthetic’. The European Copyright Society (footnote 48 of its opinion) had recommended that the Court uses such terminology in a clear way to avoid any confusion especially in view of the fact that Court’s decisions are translated into all languages of the EU. It is pleasing to see that the AG embraced this view and went even further.

As to the criterion of the author’s intention to determine a work’s originality, AG Szpunar rightly says that as long as this intention can be seen in the work, it can be taken into account. But if it is not seen in it, it is irrelevant. This is because for a work to be original, as the CJEU has held, “it is both necessary and sufficient that it reflects the personality of its author, as an expression of his or her free and creative choices” (author’s emphasis) (para. 45). So, if a court has been given evidence of the author’s intentions, it must always check whether these are indeed reflected in the work itself. He adds that an author does not need to intend to create, therefore an author’s state of mind when s/he created is irrelevant.

Building on Brompton, for the AG, all remaining factors listed by the referring courts can be taken into account to assess originality, provided the court keeps firmly in mind the criterion of originality. Therefore, if an author uses generally available shapes, the work can be original if such combination reflects the author’s free and creative choices. The AG then affirms the copyright principle of independent creation, adding that while it is true that works of applied art created independently can look similar or identical because of the constraints naturally posed to their authors, such works can still be original for copyright purposes. Finally, the exhibition of works in museums and recognition in professional circles can be elements which can confirm the originality of the work because a work which is very artistic is usually unique in the sense of reflecting the personality of the author, but it can never be neither necessary nor sufficient. Courts must check if the work is in a museum or recognised by professional circles because of its creativity not, for instance, because of its technical prowess or novelty.

Infringement

The Swedish court also asked questions about the infringement test and which factors can be taken into account.

To start with, AG Szpunar stresses again that copyright and design law are different bodies of law and that the correct application of the infringement test in the two rights is as important as the application of their protection requirements. This refers to the fact that the global impression test is used in design law but foreign to copyright law, as per the case law of the CJEU. He then reiterates the infringement test in copyright law namely that an original part, even small, of a work needs to be copied (Infopaq) and considers that the Pelhamtest of recognisability established for sound recordings can be applied to original works too. So, for him, a court should check whether “those elements that are the expression of choices reflecting the author’s personality, have been reproduced in a recognisable manner in the allegedly infringing subject matter” (para. 67). The overall impression is not sufficient and he goes further: it should not even be raised by the court, killing off the test completely in copyright law.

As to the factors courts can take into account, first, he believes that the degree of originality has no place in copyright law. For him, what the Court had said in Cofemel and Paineralso extends to infringement i.e. that the scope of protection does not depend on the degree of originality of the work. Second, “[w]here the subject matter for which protection is claimed consists of known elements of which only the arrangement is original, a reproduction of that arrangement will constitute an infringement, whereas the mere reproduction of known elements will not” (para. 71). Third, simply following the same artistic trend or current as the author of an earlier work does not constitute infringement if the creative elements of that work are not copied. Finally, he refers to the point he made previously on originality: if it is proven that the allegedly infringing work was created independently, there is no infringement and vice versa.

Comment

The protection of works of applied art by copyright is a notoriously thorny topic. The Court has had to grapple with it chiefly in Cofemel and Brompton and left several questions unclear, hence why two national courts from different Member States referred a long list of questions to the Court for clarification. AG Szpunar can be congratulated on his opinion overall. The AG’s most notable and laudable contribution is to indeed clarify fully these issues. It is now crystal clear that there is no higher or stricter requirement of originality for works of applied art, that some factors can be taken into consideration but not others to assess originality, and that neither the test of overall impression nor the degree of originality of the work can be used to assess copyright infringement.

Three important points need to be highlighted.

First, on originality, one of the interesting points the Court may pick up on is the presumption of creativity for works that the AG creates. If there is no such presumption for works of applied art, this means correspondingly that there is a presumption for all other works. This seems too sweeping a statement because works of applied art are not the only utilitarian works, software and databases for instance are too. It is debatable there should be a presumption of originality for other utilitarian works, especially for databases. In any case, creating a presumption may not be within the competence of the Court as this may be regarded as a procedural aspect and the principle of procedural autonomy applies. For a discussion see Cabay. In any case, the Court would do well to tread carefully here.

Two things must be noted on the infringement test. First, there is the point about applying the recognizability test to original works and not just sound recordings. If this means that the defendant must have taken the author’s own intellectual creation, it boils down to the Infopaq test. But the AG does not elaborate on this. This is one point where his opinion could have been clearer. But arguably, this is implicitly what he means by “those elements that are the expression of choices reflecting the author’s personality, have been reproduced in a recognisable manner in the allegedly infringing subject matter”. In Pelham, the court said recognizable to the ear, but did not lay down a specific person to do this. This is normal in copyright law as it is the judge who makes this assessment. It would be new if the Court added a specific fictitious person like in design, trade mark or patent law. The Court should not create such a person and should stick to the Infopaq test, and if it uses recognizability, it should say that the two tests are the same. Incidentally, the German legislature equated the two tests when amending article 23 of its copyright act following Pelham (cf. Explanatory Memorandum, German Parliament, 19. Wahlperiode, Drucksache 19/27426, p. 78-79). Even in the case of a non-identical reproduction, it is possible to recognise the claimant’s work, such as in a parody.

The second point is about the degree of originality of the work. Here the Court should not follow the AG. It is simply not right that a low originality work be protected as strongly as a high originality one. This would be over-rewarding little effort and unfair to those authors who have created highly original works. Arguably, it also goes against Infopaq and can be reconciled with Painer and Cofemel, i.e. the Court did not arguably say this expressly and thus this would not amount to overturning its previous decisions.

The – surely purely coincidental – timing of the opinion on the anniversary of WWII’s end thus augured well as it a victory for clarity and so a victory for authors, right holders, users, and lawyers, including judges. It is hoped that the Court will follow the AG on all points except on the degree of originality for assessing infringement. It should also be careful about the presumption of originality point. Most crucially, the Court will hopefully adopt the same clear and detailed style as the AG. Lack of clarity in its case law is the main cause of the disharmony on these issues at national level (as expounded in my forthcoming book), leaving great uncertainty and unfairness for EU citizens.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

In October 2023, several music companies (Concord, ABKCO Music & Records, Universal Music) sued Anthropic AI in the US for alleged harm to their business interests because (a) its AI chatbot, Claude, was trained with unauthorized music lyrics data  and (b) Claude’s outputs in response to user queries contained  copies of (parts of) these lyrics or derived lyrics.

On 25 March 2025, the US Northern district court of California dismissed the motion for an injunction. The judge argued that the plaintiffs had not demonstrated how using their lyrics as an input to train Claude could reduce the value of these lyrics in licensing markets or harm the reputation of the owners of rights to these lyrics. Moreover, the judge pointed out that these alleged harms relate to the unauthorized reproduction of lyrics in Claude’s outputs. However, that complaint was resolved by an agreement between Anthropic and the plaintiffs whereby Anthropic agreed to put in place guardrails that prevent Claude from reproducing (parts of) copyright-protected lyrics in its outputs. That agreement effectively prevents harm in licensing markets for lyrics but does not prevent Anthropic from using lyrics as a training input for AI models. The judge thereby confirmed that in this case the use of lyrics for training AI models does not violate US copyright law.

Meanwhile, both the plaintiffs and the defendant are confident that their views will prevail in higher courts should the case go on appeal. Indeed, this judgment is just one among many judicial steps that dozens of AI-related copyright cases will have to go through in the US. It will take several years to resolve this legal uncertainty in court.

The case is of general importance however for all copyright industries, far beyond music and lyrics, because the court stresses the underlying economic reasoning in favour of unrestricted use of copyrighted material for AI training: it does not harm the market for copyrighted outputs, unless AI models re-produce outputs that are nearly identical to training inputs. While early AI models were prone to doing so, the latest models have built-in guardrails to prevent regurgitationof inputs.

From an economic perspective, copyright grants an exclusive monopolistic right to the creator for the exploitation of the protected work. Monopolies are, almost by definition, inefficient from a societal point of view. However, in the case of copyright, they constitute an incentive to invest in new creative products that are beneficial for society. The scope of copyright protection should ensure that these dynamic innovation benefits exceed the welfare losses for society from granting an exclusive copyright. A key test for this economic criterion is the impact of an extension or reduction in scope on the supply of new creative products. The California judgment in the Anthropic case captured that economic reasoning very well.

There is no evidence so far that use of copyright protected materials as inputs for AI model training would harm the market for these materials. Authors can continue to sell their creative products in their usual outlet channels. Of course, AI may increase competition from AI-generated or hybrid products. But as long as these are not (nearly) identical reproductions of training inputs in model outputs, that is acceptable. In line with economic reasoning, the judge distinguished between possible harm from violation of reproduction rights in AI model inputs and outputs and accepted that only the latter could be harmful to the market for licensing of lyrics.

Just like firms, authors do not like competition. But consumers benefit from that competition. Evidence shows that authors tend to avoid publishing their copyrighted content on platforms that are known to use this for AI training. But that does not imply a reduction in production. However, there is evidence that limiting the scope of Text and Data Mining exception to copyright, is harmful for innovation in society. A third of all websites are already implementing copyright opt-out protocols, such as robot.txt, in line with the EU Copyright in the Digital Single Market Directive (CDSM Directive) Art 4(3). That reduces the volume of AI training data and the quality of AI models. It has a negative economic impact on AI-driven innovation across the entire economy, far beyond media industries, which account for less than four percent of GDP.

Copyright holders may of course claim that whatever US courts decide with regard to copyright and AI has no bearing on EU AI markets. They may feel protected by extraterritoriality claims in the AI Act that, though controversial, may give European authors a feeling of protection against liberal US interpretations of copyright in AI model applications. This may result in a false sense of economic isolationism. Foundational generative AI or Large Language Models, such as OpenAI’s ChatGPT, Meta’s Llama and Google’s Gemini, are all developed and trained in the US, but widely used in the EU. Very few foundational models, and certainly no current technology frontier models, are trained in the EU. The absence of suitable computing infrastructure in the EU, and the fact that all big tech firms are US-based, has a lot to do with this. But uncertainty and risks surrounding the EU AI copyright regime are an important factor too. If the EU wants to strengthen its sovereignty in the AI supply chain and promote EU home-grown AI models, it should take copyright economics seriously.

Rightsholders may reserve their rights with opt-outs in the hope that it generates licensing revenue. This may work for a few very large publishers who  have indeed signed licensing agreements with some AI model developers.  It is unlikely to work for smaller publishers. Trying to do so with millions of unidentified web publishers would run into unsurmountable licensing transaction costs. Collective licensing only displaces that problem from AI developers to intermediary agencies; it does not solve that problem.  This results in biased AI training datasets, limited to a few large publishers only and  goes against the anti-bias requirements imposed by Recitals 70, 110 and 156, and Art 10 § 5, of the AI Act.

Creative industries already benefit substantially from the use of AI models for the production of content. At the same time, they claim that these models harm them. Their schizophrenic mindset is perhaps best captured in copyright measure 2.3(5) in the proposed AI Code of Practice. It reiterates that AI model developers should respect publishers’ opt-outs from the use of their webpage content for AI model training. However, they should ensure that the same content is picked up by their search engines. Webpage publishers want their content to be found by users of AI algorithms in search engines, but not by AI model algorithms that deliver a more reasoned response to user queries. In other words, EU copyright allows humans to learn from their content, but not more efficient machines that work for humans.

This policy stance on (the economics of) AI copyright risks marginalizing the EU even further in global AI markets and services, both with regard to model training and use, especially if US jurisprudence would move towards fair use and transformative use of copyrighted content in AI model training. Some big tech firms are already withholding their most advanced AI models and services from the EU market, so far mostly related to legal uncertainty about the use of personal data for training of AI models. Even if, miraculously, EU extra-territoriality claims with regard to copyright in AI would be admissible, they will not protect it from falling into the trap of low-quality and underperforming AI services.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

There is a bit of excitement in copyright circles about the first case referred to the CJEU that directly addresses the intersection of artificial intelligence (AI) and the EU copyright framework. The request for a preliminary ruling — Like Company v Google (C-250/25) — originates from the Budapest Capital Regional Court (Budapest Környéki Törvényszék) and involves a dispute between Like Company, a publisher and operator of various online news portals, and Google, in its capacity as the operator of the Bard (now Gemini) chatbot.

Like Company claims that responses provided by Bard, in reply to requests to summarize the content of a specific web page, infringe its rights under the relevant national and EU legislation (copyright and/or the neighbouring right for press publishers), as the response constitutes an unauthorized communication to the public. Whether chatbot answers that summarize publicly available information protected by the press publishers’ right constitute a communication to the public indeed seems like an interesting new question for the CJEU to answer^[1] — and one I’ll gladly leave to more qualified people to opine on.

Instead, I will focus on another — somewhat problematic — aspect of the referral: it appears to misrepresent some of the underlying technical processes, which has led the court (and some commentators) to frame the central issue as one concerning the legality of training AI models on publicly available content. In the second and third questions referred to the CJEU, the Budapest Capital Regional Court asks (emphasis mine):

• Must Article 15(1) of Directive 2019/790 and Article 2 of Directive 2001/29 be interpreted as meaning that the process of training an LLM-based chatbot constitutes an instance of reproduction, where that LLM is built on the basis of the observation and matching of patterns, making it possible for the model to learn to recognise linguistic patterns?
• If the answer to the second question referred is in the affirmative, does such reproduction of lawfully accessible works fall within the exception provided for in Article 4 of Directive 2019/790, which ensures free use for the purposes of text and data mining?

And while the latter question is indeed the billion-euro question when it comes to the applicability of the EU copyright framework to AI training — and one that the CJEU will likely have to answer at some point — the connection between this issue and the facts at hand in Like Company v Google seems spurious at best. Yes, there is little doubt that Bard (now Gemini) is based on an AI model trained on large amounts of copyright-protected (and non-protected) material sourced from the public internet. But based on the facts as established by the Budapest District Court, it seems highly improbable that the alleged infringement results from reproductions made during the training of the AI model that the chatbot in question was based on.

The underlying facts that gave rise to the dispute are presented in points 7 and 8 of the “succinct presentation of the facts and procedure in the main proceedings” section of the referral document:

1. An article appeared on one of the applicant’s protected online press publications (balatonkornyeke.hu) stating that Kozsó, a well-known Hungarian singer, had not given up on his dream of putting dolphins in an aquarium next to Hungary’s largest lake, Lake Balaton. That article also made reference to other online press publications belonging to the applicant, reporting on the hospitalisation of Kozsó, his interests, the fact that he had served a custodial sentence in the United States and also a fine he had received for electricity theft.
2. In response to the question ‘Can you provide a summary in Hungarian of the online press publication that appeared on balatonkornyeke.hu regarding Kozsó’s plan to introduce dolphins into the lake?’, the defendant’s chatbot provided a detailed response which included a summary of the information appearing in the news media belonging to the applicant.

Dolphins in Lake Balaton

The description in point 7 makes it very likely that the article at issue is Kozso nem adja fel: továbbra is delfineket szeretne a Balatonhoz telepíteni a népszerű énekes(which translates to “Kozso doesn’t give up: the popular singer still wants to introduce dolphins to Lake Balaton”) , published on 21 July 2023.

It is the description of the actual mechanics of the case in point 8 that makes it clear this case is not about the training of AI models, but about something else entirely. What seems to have occurred is that a user — with prior knowledge of the article in question — directed the chatbot to provide a summary by referencing the domain name of the publication where the article was published and providing enough contextual information to identify the specific article. In response, the chatbot (an LLM) accessed the content of the website and generated a summary of the text found there.

Given the close temporal proximity between the publication of the article (21 July 2023) and the period for which infringement is alleged (13 June 2023 to 7 February 2024), it seems highly unlikely that the underlying model had been trained on the content of that specific article^[2],[3]. Instead, it appears almost certain that the already trained model used the live content of the website as input, and then operated on it to produce the requested summary. This interpretation is also supported by the defendant’s explanation, summarized in point 23: “In order to collect data, [the chatbot] uses the Google Search database, and, in its response, it is able to display a modified version of an article, if the user has already provided the original version of the article in his or her instructions.” In other words, upon receiving the prompt, the chatbot searched the Google Search index for content from the referenced website and then produced a summary based on that content – a type of process often referred to as Retrieval Augmented Generation (RAG).

While such interactions with chatbots — and their ability to summarize websites on demand — may still seem novel, the overall process is not. Attentive readers may notice that the translation of the article provided above via Google Translate is the result of an analogous process. Given a pointer to the article (in this case, the URL), a service operated by Google (Google Translate) uses the content of the website as input for an AI model, which then transforms it into the requested output (an English translation). The only substantive difference is that, in the translation case, Google goes to great lengths to preserve the overall structure and context of the original website^[4], whereas in the summary case, the output is presented within the chatbot interface, which bears little or no relation to the source website.

Based on all of this it seems safe to conclude that the case as referred to the CJEU does not in fact deal with issues related to the training of AI models but rather with issues arising from their use. This distinction is important for at least two reasons: On a practical level there is a real danger of arriving at conclusions that can limit the freedom of individual users to interact with publicly available content based on mistaken understanding of the underlying technology. And on a more general level it seems important that decisions related to the applicability of the TDM exception to AI training will be made based on a case that actually involves AI training. As I have shown above that is almost certainly not the case here at least not in the terms described by the court.

^[1] The article in question at the center of the dispute certainly makes a great addition to the eclectic CJEU case law on communication to the public.

^[2] Training large AI models such as bard generally takes months and they commonly have knowledge-cut off dates that are well before they are deployed.

^[3] Note that there is a slight inconsistency here between the publication date and the presentation of facts that alleges that the making available to the public occurred between 13 June 2023 and 7 February 2024. The most likely explanation is that one of the dates is not correct.

^[4] This includes the provision of a URL that makes great efforts to appear as if the content is hosted on the original website, but that at closer inspection reveals itself as a URL fully controlled by Google: translate.goog

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

The ongoing public discourse on Sustainability – meant here not in its plain-language meaning but rather in its evolving definition in law- and policy-making (see e.g. Verschuuen) – pivots, to a significant extent, on the need to promote better models of consumption and production. This need to optimize the exploitation of natural resources and reduce waste finds an important ally in the so-called repair culture. The EU has moved important steps forward in fostering the repair of products as part of its Green Deal and Circular Economy Action Plan. It did so mainly by intervening in its consumer protection legal framework, with the aim to oblige some manufacturers to repair their products as well as to ensure a fair EU market for independent spare parts and repair service providers (see the 2024 EU Repair Directive).

The intention to foster a culture of repair has also permeated the EU IP legal framework. The recent EU Design Law reform package is a glaring example, as it introduced the so-called repair clause, taking the cue from the famous CJEU ruling in Acacia and fully harmonizing the legal defence of repair for spare parts manufacturers (see Article 20a EU Design Regulation and Article 19 EU Design Directive).

What the EU Design reform performed is an utterly necessary legislative finetuning. In fact, Article 5(6) Repair Directive prohibits manufacturers from impeding the repair of their products by way of software or hardware techniques, contractual clauses, or opposing the use of spare parts. Nevertheless, this provision applies with no prejudice to the protection of IP rights. This might seem to suggest that the EU legislator leaves open the possibility of IP legal protection to serve as a lawful ground to impede repairability.

The possible interpretations on this point are two: either the EU legislator underestimates the potential of IP legal protection in blocking repair activities, or the Repair Directive relies on the ability of the EU IP legal framework to make itself compatible with the EU Circular Economy goals, thus avoiding contradicting legislation. The newly harmonized repair clause in EU Design Law supports this latter approach. By clarifying the conditions under which spare parts manufacturers do not infringe design rights, it creates systematic cohesiveness with the obligation in Article 5(6) EU Repair Directive. EU copyright law did not perform any specific finetuning with the rise of the repair culture, despite its relevance being significant. Suffice it to think of instruction manuals, original and customized spare parts, and software components as classic examples of subject matter that can easily qualify for copyright protection. Moreover, the open question of the copyrightability of product design further strengthens the need to take copyright protection into account (see, among others, Cofemel, Brompton, and pending case Konektra).

This means that, potentially, the production and commercialization of a spare part may not amount to design infringement, but the question might arise as to copyright violation. This is due, in particular, to two main legal considerations.

First, the only partial harmonization of the copyright exception dedicated to the repair of equipment. Art.5(3)(l) of the InfoSoc Directive, like many other parts of the same provision, is a copyright exception that Member States can choose to implement, but are not obliged to. In light of ReCreating Europe’s mapping of copyright flexibilities, Rosborough’s analysis and our recollection, across the EU, only 10 Member States did so.

Second, more strictly related to the digital and software dimensions of repairing activities, the Computer Program Directive does not address any specific needs underlying the repairability of products. Besides not introducing a mandatory exception to use software for repair purposes, it also does not provide rules on Technological Protection Measures (TPMs) that match with the obligation stemming from the Repair Directive not to block repairers by way of software, hardware, or contractual arrangements.

In light of the above, it is worth asking whether EU copyright law will be next to finetune its provisions with the need to foster the EU repair culture – and if so, how. Several relevant experiences across the world might serve as inspiration. As Rosborough explains, the legislative developments often labelled under the motto “right to repair” come in different shapes, ranging from the introduction in legal systems of rights or legal defences for individual consumers and independent repairers, up to the establishment of new obligations for original manufacturers to market repairable products.

Among the most notorious examples is the Canadian legal system, whose reform specifically targeted the Copyright Act and allowed the circumvention of TPMs for repairability and interoperability purposes (see An Act to amend the Copyright Act (diagnosis, maintenance and repair) and An Act to amend the Copyright Act (interoperability), both of 2024). Similarly, in the US, where right to repair laws have been introduced in every State, federal copyright law (in particular, the Digital Millennium Copyright Act) allows for temporary exceptions to the prohibition of circumvention of TPMs for the purpose of repairing products (for an on-point critical analysis, see e.g. Rimmer).

As the experience in these countries suggests, the need for a finetuned copyright legal framework is expected to also come to the surface in the EU, either in the legislative debate due to the growing focus on Sustainability regulation or in judicial settings, as repair becomes an ever more cross-border market practice. Besides the imperative to take the opportunity to build systematic cohesiveness of the EU legal system, this aspect could be, in our view, of tremendous impact in stimulating the policy and doctrinal debate  on possible ways forward – creating momentum around the effectiveness of regulating the interplay between copyrights and repair culture statutorily or judicially, on the language to be used, on the need for redrafting copyright exceptions and the three-step-test (see, among others, the proposal for IP flexibilities for any uses extending the lifespan of products by Pihlajarinne), and, importantly, on the need to modernize the copyright legal system in light of current market practices and EU values and objectives.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

Introduction

Not long ago, artificial intelligence (“AI”) was a concept brought to life by human actors – whether through Scarlett Johansson’s voice in Her (2013) or as Alicia Vikander’s eerie humanoid presence in Ex Machina (2014). Today, the roles have reversed: it is AI that is creating on-screen performances that appear convincingly human.

From de-aging and re-aging to posthumous digital replicas and fully synthetic background performers, generative AI offers the entertainment industry new creative possibilities and potentially cost-saving efficiencies in (post)production. For screen and voice actors and actresses (collectively “performers”), AI can come to the rescue when illness or unforeseen circumstances prevent them from completing a project – potentially avoiding a full recasting. Yet fears about the erosion of their creative control, and ultimately their livelihoods, clearly prevail.

These concerns fueled the four-month strike by the U.S. actors’ union SAG-AFTRA in 2023, when AI became a key issue in the negotiations over the renewal of its agreement with the production companies’ association AMPTP. In December 2023, the new contract (“SAG-AFTRA Agreement”) was ratified, introducing detailed provisions governing AI in film and television production.[1] In February 2025, the German federal actors’ union BFFS and the public sector trade union ver.di reached an agreement with the producers’ alliance (Produktionsallianz) on the use of generative AI in film production (“BFFS Agreement”), effective March 1. While this agreement was reached without the high-profile strikes seen in Hollywood, it represents an equally significant milestone for the film industry.

This two-part article examines the BFFS Agreement and compares its key provisions to the SAG-AFTRA Agreement, highlighting similarities and differences in ensuring performers’ participation and control over the use of AI. Part 1 explored statutory protection vs industry agreements, regulated AI use cases, and compensation for AI replacement. This second part looks at consent requirements and post-mortem AI-modified performances and considers the future outlook in this field.

IV. Consent Requirements

Both the BFFS and the SAG-AFTRA Agreement recognize that no consent is required where the use of performance or likeness is legally permitted. Under German law, this includes cases where other fundamental rights (e.g., freedom of art) prevail or copyright exceptions (e.g., quotation, pastiche) apply. Under the SAG-AFTRA Agreement, this concerns First Amendment uses (e.g., comment, criticism, satire, parody, docudrama). Otherwise, the following consent requirements apply.

1. Digital replicas

Only very narrow exceptions to the requirement of the performer’s explicit consent apply for digital replicas.

(a) General rule: Consent required

The BFFS Agreement reflects the statutory requirement of consent for any creation, reproduction, distribution and public display of a digital replica – regardless of whether it is created during the performer’s engagement for a specific production or based on pre-existing recordings, and of the performer’s physical participation.

The SAG-AFTRA Agreement follows the same broad approach of a general need for clear and conspicuous consent for any kind of digital replica. Some details differ due to the distinction between digital replicas of background actors and those of performers, with the latter further divided into employment-based (created in connection with the performer’s employment on a film) and independently created (there is no employment for the performer on the film in which the digital replica is used).

While consent and granting of rights under the BFFS Agreement may be included in the general cast agreement, the SAG-AFTRA Agreement requires a separately signed declaration.

(b) Exceptions to the need for consent

The BFFS Agreement provides that the performer may not, contrary to good faith, refuse to consent to the use of a digital replica or partial digital embodiment for (stunt) scenes in which she is shown or doubled in “dangerous situations”. But the need for consent is only waived fully if the use of a digital replica or partial digital embodiment does not “significantly exceed the scope of the contractually agreed real participation” of the performer and the performer is unable to perform, for example due to an injury. The SAG-AFTRA Agreement is broader: no consent is required if the replica use leaves the photography or soundtrack substantially as scripted, performed and/or recorded, regardless of whether the performer was able to perform in person.

(c) Use in subsequent productions

Additional consent requirements apply when a replica is used outside the production for which the performer was originally engaged. The BFFS Agreement mandates consent in writing (which is not generally mandatory for digital replica, but always advisable) and in a separate agreement with specified use of the replica (production, role, etc.) and separate remuneration.

Consent must generally be obtained when the digital replica is used in a subsequent production. It may be included in the cast agreement for the initial production only if the performer is already engaged for the later project (e.g., in a series, prequel, sequel, remake). In general, digital replicas may only be used in other projects if the performer is re-engaged, except for when the performer is unavailable due to other obligations, illness, or death, or if the parties conclude an individual agreement for compensation.

In essence, this mirrors the SAG-AFTRA Agreement’s approach to digital replicas of performers.[2] For background actors, however, consent must always be obtained at the time of use, and use of digital replicas in other projects is prohibited if this is done to circumvent a re-engagement.

2. Digital modifications

Material digital modification of the age, physical stature, appearance and/or voice of a performance using generative AI require express written consent of the performer to the specific modification (email suffices under the BFFS Agreement; the SAG-AFTRA Agreement requires a separately signed statement). The following exceptions apply under the BFFS Agreement, all of which remarkably similar to the carve-outs in the SAG-AFTRA Agreement:

(a) No consent required: Script-compliant modifications

No explicit consent is needed for modifications that remains substantially faithful to the script or the performer’s role. Since the agreement aims to ensure sufficient control and protection for performers, this exception is likely limited to minor AI-enabled changes. As an example of a substantial departure from the script going beyond this exception, the BFFS Agreement mentions “(visual) reversals”, e.g., reskinning by changing a dark-skinned into a light-skinned person or vice versa.

(b) No consent required: Customary post-production editing

Within the statutory protection against distortion, industry-standard editing and adjustments are permitted without separate consent. This privileging is reminiscent of the EU AI Regulation, which exempts assistive AI standard edits from the provider’s obligation to ensure labelling of synthetic content, Art. 50(2).

The BFFS Agreement lists typical post-production measures, including editing, cutting and other changes to repair or improve the quality of the recorded material (e.g., color grading), in costume, for timing and speed, for continuity, noise reduction, intelligibility, length, sound, for VFX effects and filters, but also for changes made to comply with ratings (e.g., for youth protection) and minor adjustments to dialog, storylines, legal requirements and industry practices. However, any deviation from the script or role should be handled with great care and, when in doubt, covered by explicit consent.

(c) No consent required: Dubbing

Digital modifications may also not require separate consent if they are made for post-synchronization to dub the film into another language, including adaptations of the dialog or recorded performance for distribution in certain license markets and adjustments made to the performance (i.e., use of a double or lips/face/body/voice changes).

3. Artificial performers

For artificial performers, the BFFS Agreement reiterates the legal status quo: consent is required if a human performer is recognizable for acquaintances in the artificial performer. It also affirms the value of human performance and acknowledges AI’s potential impact on employment, echoing the language of the SAG-AFTRA Agreement. However, the SAG-AFTRA Agreement already requires producers to give notice and bargain in good faith if a synthetic performer is to replace a human one.

V. Post-mortem AI-modified performances

The SAG-AFTRA Agreement treats the effect of a performer’s or background actor’s death on the consent given during her lifetime as a key issues, again due to fragmented U.S. state law and the lack of a federal framework.[3] Such consent remains valid after death unless explicitly limited. Where consent must be given after the performer’s death (e.g., for replica use in another production or post-mortem digital alterations), it may be granted by the estate or, if none exists, by the union.

The BFFS Agreement addresses a performer’s death only with respect to the use of a digital replica in subsequent productions: consent given during the lifetime in the context of the initial engagement that covered later uses remains valid after death. Apart from that, general copyright and personality rights apply to post-mortem uses, meaning any use after the death of the performer not adequately contractually covered during the lifetime must be authorized by the successor in right.

VI. Conclusion and Outlook: More similarities than differences

The BFFS and SAG-AFTRA Agreements both mark significant landmarks in the industry regulation of AI in film production, each shaped by their legal framework. Since German law already provides strong protection for performers in their performances and likenesses, the BFFS Agreement focuses on practical implementation and establishing industry-wide agreed modalities. This makes the agreement no less important, as it defines clear standards and a fair balance of interests.

Both agreements address current key use cases of generative AI output, the BFFS Agreement even partial digital embodiments, whereas the SAG-AFTRA Agreement does so only in the context of synthetic performers with recognizable key human facial features. Consent and compensation are key principles in both agreements, with similar approaches to the scope of consent (generally project-specific, with exceptions if the performer is re-engaged), its redundancy for certain digital modifications (script-faithful changes; standard post-production editing; dubbing), and the calculation of compensation for AI replacements (fictitious shooting days). Notably, the SAG-AFTRA Agreement even waives consent if a digital replica leaves the photography or soundtrack substantially intact.

A key difference lies in post-mortem use of AI-modified performances: the BFFS Agreement follows an opt-in model, while the SAG-AFTRA Agreement applies an opt-out approach (lifetime consent remains valid unless limited).

Both agreements foresee regular reviews. The SAG-AFTRA Agreement calls for semi-annual meetings with each producer to discuss generative AI use and bias mitigation. The agreement’s 2.5-year term (until end of June 2026) will prompt renegotiations by January 2026. The BFFS Agreement also provides for biannual joint reviews and earlier termination than the general collective agreement for flexibility.

To conclude, the BFFS Agreement reflects the core components and main compromises of the SAG-AFTRA Agreement. As film speaks a universal language, a fully unified global industry standard would be desirable. So far, Hollywood has taken the lead in shaping such a standard.

—————————————————————————————————————

[1] Since then, SAG-AFTRA has concluded other AI-specific agreements, for example with a AI voice technology company.

[2] The SAG-AFTRA Agreement only allows replica use without re-engagement if the performer is deceased. But it also recognizes “independently created digital replica” – uses for productions where the performer is not employed but has consented against bargaining – similar to the BFFS Agreement’s exception permitting use without re-engagement through individual agreement against payment.

[3] California just passed a SAG-AFTRA-supported law in August 2024 that requires the estate’s consent for the creation and use of digital replicas of deceased personalities.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

I. Introduction

Not long ago, artificial intelligence (“AI”) was a concept brought to life by human actors – whether through Scarlett Johansson’s voice in Her (2013) or as Alicia Vikander’s eerie humanoid presence in Ex Machina (2014). Today, the roles have reversed: it is AI that is creating on-screen performances that appear convincingly human.

From de-aging and re-aging to posthumous digital replicas and fully synthetic background performers, generative AI offers the entertainment industry new creative possibilities and potentially cost-saving efficiencies in (post)production. For screen and voice actors and actresses (collectively “performers”), AI can come to the rescue when illness or unforeseen circumstances prevent them from completing a project – potentially avoiding a full recasting. Yet fears about the erosion of their creative control, and ultimately their livelihoods, clearly prevail.

These concerns fueled the four-month strike by the U.S. actors’ union SAG-AFTRA in 2023, when AI became a key issue in the negotiations over the renewal of its agreement with the production companies’ association AMPTP. In December 2023, the new contract (“SAG-AFTRA Agreement”) was ratified, introducing detailed provisions governing AI in film and television production.[1] In February 2025, the German federal actors’ union BFFS and the public sector trade union ver.di reached an agreement with the producers’ alliance (Produktionsallianz) on the use of generative AI in film production (“BFFS Agreement”), effective March 1. While this agreement was reached without the high-profile strikes seen in Hollywood, it represents an equally significant milestone for the film industry.

This two-part article examines the BFFS Agreement and compares its key provisions to the SAG-AFTRA Agreement, highlighting similarities and differences in ensuring performers’ participation and control over the use of AI. One thing is clear from the outset: both agreements share the same basic approach to AI – not banning it, but using it responsibly is the future of filmmaking.

II. Statutory Protection vs. Industry Agreements

Under German ancillary copyright protection, any recording of a performance and its copyright-relevant use (e.g., reproduction, communication to the public), requires the performer’s consent and appropriate remuneration.[2] This, and the moral rights protection against gross distortions, is rooted in international copyright law and applies equally to uses involving generative AI. The German constitution embeds a general right of personality, which, together with specific law, gives performers additional control over their image and likeness.[3] Unauthorized and any defamatory use of a performer’s likeness, image, or other personality traits such as voice, style or mannerism is prohibited. If such uses are not covered in cast agreements, studios and producers may face liability for copyright infringement and violation of personality and data protection rights.

Given this strong legal protection of performers against AI-driven modifications and imitations of their performances and likeness, the BFFS Agreement functions primarily as a framework for standardized implementation. In contrast, performers in the U.S. operate in a legal landscape without a strong federal protection, which is why collective bargaining agreements are crucial to establishing adequate protections and ensuring a level playing field between performers and studios.

III. Regulated AI Use Cases

The BFFS Agreement addresses generative AI, but not all applications of the technology that might technically qualify as such fall within its scope. Similar to the SAG-AFTRA Agreement, “conventional AI technology applications” already used by film producers are excluded, such as CGI and audiovisual effects, but also the recreation and alteration of voices and images. The agreement focuses on regulating new production capabilities enabled by AI, rather than applying to all possible uses of generative AI. Four specific use cases of generative AI are recognized:

• Digital modifications: Changes and alterations of the image or performance (visual and/or acoustic) made with generative AI.

• Digital replica: AI replica of a performer’s voice or image that gives the impression that the recreation is the performer herself and is used to virtually place the performer in scenes filmed without her participation.

• Partial digital embodiments: Modifications and alterations of only parts of the image or the (visual and/or acoustic) performance made with generative AI to transfer individual body parts such as the face to other people (e.g., stunt or nude doubles) or to replace the performance with AI-generated elements (e.g., virtuoso piano-playing hands). This process creates human-human or human-AI hybrid characters.

• Artificial performers: Digital “beings” created with generative AI by mixing recordings of multiple performers into a new entity that is no longer recognizable as a particular identifiable person, but is intended to, or does, give the impression of being a natural person. The SAG-AFTRA Agreement refers to “synthetic performers” in this context.

The SAG-AFTRA Agreement covers digital alterations and digital replica, but not partial digital embodiment. This difference may be due to the strong statutory protection in Germany, which can be broken down to individual features of the performer. The SAG-AFTRA Agreement does, however, offer a limited protection of physical characteristics when it comes to synthetic performers. Even though synthetic performers are defined as ultimately not recognizable as a particular person, producers must obtain consent to use a performer’s name and key facial features (eyes, nose, ears and/or mouth) in AI prompts to generate a synthetic performer with a recognizable principal facial feature. The agreement over the use is left to the parties. This remains the only explicit reference to the input level of AI. The BFFS Agreement also mentions this only as an afterthought to the definition of digital motion capture (the capturing a performer’s facial and/or body movements to create a recording of a model for digital replica or partial digital embodiments), noting that digital data generated there “can also be fed into a generative AI process if necessary”.

IV. Compensation for AI Replacement and Consent

In general, a fee must be specifically agreed upon for the performer’s consent (see V.) to use the digital replica or digital partial embodiment. Both agreements base this on “fictitious shooting days”, the days that would have been required for the production if the performer had personally acted instead of the AI-replacement. The BFFS Agreement calls for objective criteria such as the length of such a scene with human involvement as estimated in good faith by the producer (“stopped script length”), but keeps this under review, as producers prefer a “screentime approach” based on the digital replica’s actual visibility and audibility in the final product.

Since AI replacements are to be compensated like an own performance, a separate fee need not be agreed upon if (i) the performer is already compensated for the production and the use of the digital replica or digital partial embodiment does not significantly exceed the scope of the performer’s contractually agreed-upon participation, or (ii) the AI replacement is used in a scene in which the performer appears anyway, for which the BFFS Agreement acknowledges the exceptions of double and additional roles. Work for digital motion capture or creating a digital replica counts as shooting days and must be compensated accordingly.

Remuneration for fictitious shooting days is to be based on the general film workers collective agreement, which the BFFS Agreement’s parties have also recently revised. Due to the prominent role of residuals in the Hollywood entertainment industry and their essential role in ensuring fair compensation for ongoing film exploitation, SAG-AFTRA secured that the residuals for AI-replaced performances match those for human performances.

Part 2 of this post will explore consent requirements and post-mortem AI-modified performances.

—————————————————————————————————————-

[1] Since then, SAG-AFTRA has concluded other AI-specific agreements, for example with a AI voice technology company.

[2] Sect. 73, 77, 78, 79(2a), 32 German Copyright Act (Urheberrechtsgesetz, UrhG).

[3] Art. 2(1), Art. 1(1) German Constitution (Grundgesetz); Sect. 22 Art Copyright Act (Kunsturhebergesetz, KUG).

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

The debate on whether works protected by copyright can be used for the training of artificial intelligence (AI) has reached India. While dozens of US District Courts are currently grappling with the question of whether AI training with protected works constitutes fair use, the UK High Court is largely grappling with jurisdictional questions, and EU courts are mainly concerned with the modalities of rights reservations (see for overviews here, as well as here and here), it is now the High Court of Delhi’s turn. The essential question in this first Indian AI case is whether the use of works for training purposes is covered by an exception, or whether AI developers must obtain authorisation for the works used to train their AI systems.

Background:

In November 2024, the news agency Asian News International Media Private Limited (ANI) filed a case before the High Court of Delhi, India (Ani Media Pvt Ltd vs Open AI Inc & Anr. [CS(COMM) 1028/2024]). ANI alleged that Open AI had used ANI’s content to train its Large Language Model (LLM), particularly Open AI’s ChatGPT, without obtaining adequate permission from ANI for such usage. ANI contends that some of its material was accessible only to its subscribers and that Open AI has no authorisation to use the openly available and paywalled materials which were republished by ANI’s subscribers. Furthermore,  ANI claims that Open AI had falsely attributed false news to the agency, damaging its reputation/ spreading misinformation.

In its application, ANI sought an ex parte and interim injunction on two matters. First, that Open AI or any person acting on Open AI’s behalf be restricted from ‘storing, publishing, reproducing or in any manner using, including through the ChatGPT model, the copyrighted work of ANI or any other original works of ANI.’ And, second, that ‘Open AI be directed to disable access of ChatGPT to ANI’s works published anywhere by ANI or its subscribers.’ 

Open AI submitted that content accessible on ‘www.aninews.in’ had already been blocklisted in October 2024 and that the domain will be excluded from any future training of Open AI.  In its order dated 19 November 2024, the High Court of Delhi has set up the following questions for consideration:

1. Whether the storage by Open AI of ANI’s data (which is in the nature of news and is claimed to be protected under the Copyright Act, 1957) for training its software i.e., ChatGPT, would amount to infringement of plaintiff’s copyright?
2. Whether the use by Open AI of ANI’s copyrighted data in order to generate responses for its users, would amount to infringement of ANI’s copyright?
3. Whether Open AI’s use of ANI’s copyrighted data qualifies as ‘fair use’ in terms of Section 52 of the Copyright Act, 1957?
4. Whether the Courts in India have jurisdiction to entertain the present lawsuit considering that the servers of the defendants are located in the United States of America?

Arguments of Amici Curiae

To help answer these questions, the Court invited submissions from two Amici Curiae (Prof. Dr. Arul Scaria and Advocate Adarsh Ramanujan). The Amici made oral submissions during two hearings on 21 February and 10 March. Both argued that ANI must establish that their content is protected by copyright and that it is the lawful owner of that content,  and neither Amicus seems to contest that the acts of OpenAI engaged the reproduction right under section 14(a)(i) of the Indian Copyright Act, 1957.  However, their appreciation of the applicability of the statutory exceptions to the various stages of AI training differ significantly.

As a preliminary point, apart from the questions on the interpretation of substantive Indian copyright law, OpenAI challenges the High Court of Delhi’s jurisdiction to decide on the matter. It advances the argument that none of the relevant acts have been performed in India, a strategy that has also been adopted by Open AI in the Getty Images v Stability AI litigation in the UK. Neither Amici seems to agree with that contestation. Although the relevant acts of alleged infringement took place outside India, both argue that, according to Section 62 of the Indian Copyright Act, 1957, a suit concerning copyright infringement can be instituted in the court where the plaintiff resides or carries on business. Since ANI has its place of business in New Delhi (which is so far undisputed in these proceedings), the High Court of Delhi would have jurisdiction to hear this matter.  The Judge on the matter mentioned that he would not deal with the jurisdiction issue as a preliminary issue and would hear arguments on both merits and jurisdiction.

On the substance, it seems that the Amici did not engage in a detailed analysis of the restricted acts under Section 14 (‘Exclusive rights of reproduction vested with the Copyright owner’) in connection with Section 51 of the Act (‘Acts which amount to the infringement of copyright’). The majority of the arguments advanced focus on the question of whether such acts can be justified based on an exception under Section 52 of the Act.

Section 52 of the Indian Copyright Act, 1957 provides for certain exceptions to the exclusive rights and follows a ‘hybrid’ system of exceptions. Section 52(1)(a) provides three ‘fair dealing’ exceptions for private or personal use, which expressly include uses for research, for criticism or review, and for the reporting of current events and current affairs. Section 52 further contains a series of other specific statutory exemptions. However, none of the exceptions listed in section 52 expressly provide for the use of works for TDM, similar to articles 3 and 4 of the EU CDSM Directive, or as in the UK for computational analysis under s. 29A of the Copyright, Designs and Patents Act 1988.

In the absence of an express exception, the Amici discuss whether and how dealing with protected subject matter can be accommodated within Indian copyright law. More concretely, both submissions discuss whether reproduction for the creation of training datasets and the training itself fall under the exception for private and personal use, including for research purposes, under Section 52 (1)(a)(i) of the Act.

The submissions made by the first Amicus, Arul Scaria, suggests that the extraction of information for  purposes of AI training constitutes a non-expressive use of copyrighted works. In his oral submissions he suggests that a machine learning process is similar to the human learning process and that therefore the relevant exception under Section 52 would apply to human as well as machine learning. He advances the argument that learning is permissible under the current framework of Indian Copyright law because the AI system is trained by ‘learning’ the ingested materials. In addition, AI applications assist individuals with learning and research and storage for such purposes is also permissible under the Indian Copyright Law.  Finally, Scaria proposes that exceptions under Section 52 apply to all types of use, including uses by commercial providers of AI systems.

The second Amicus, Advocate Adarsh Ramanujan, argues that LLM training can be divided into three parts: collection of raw data, tokenisation of the collected data, and training of the model, a distinction the first Amicus had not made. He agreed with the first Amicus only to the extent that tokenising and vectorising of the collected data constitutes a non-expressive use which does not reproduce the original expression. Therefore, this stage would not constitute copyright infringement. However, the other stages (collection of raw data and training of the model) involved expressive use, which amounted to infringement. He stated that collecting and storing publicly accessible data amounted to reproduction under Section 14(a)(i) of the Act and therefore comes within the scope of infringement prescribed under Section 51 of the Act. Ramanujan seemed sceptical that any of the narrowly formulated specific exceptions listed under Section 52(1) apply to machine learning, but it would eventually be Open AI’s onus to demonstrate that the relevant acts are covered under Section 52(1).

ANI’s arguments

ANI’s lawyer argued (in part) before the High Court of Delhi on 10 March and 18 March. Building on Ramanujan’s argument which separates the training process into three stages he stated that infringement occurred at all stages of the training process as the vectorisation process resulted in an adaptation (Section 14(a)(vi) of the Act) of ANI’s work.  In addition to the infringements at the three stages of the training process, further infringement occurred at the output stage. Furthermore, ANI, as the copyright owner, had an exclusive right to use the work and any breach of that exclusive right amounts to infringement under Section 51 of the Act. These infringements cannot be justified, since Section 52 provides for an exhaustive list of instances in which prima facie infringing uses do not require authorisation, and no further permitted uses could be read into the statute apart from the ones that are expressly listed.

Comment

The outcome of the pending case before the High Court of Delhi will carry a certain significance. Whilst the written submissions of the Amicus remain unpublished, the reports of the hearings foreshadow an intense proceeding with high stakes. Beyond the issues discussed in this post, the Amici have also alluded to the question of opt-outs and filtering of generated outputs, neither of which have a statutory basis in the Indian Copyright Act. Therefore, it can reasonably be expected that the High Court of Delhi will focus on the interplay of exclusive rights and permitted uses.

In the absence of a clearly applicable exception, the answer to the question of whether the use of works for AI training purposes is lawful will determine whether India offers a tech-friendly legal copyright framework. A negative answer might induce the government to take legislative action to address an obvious lacuna in Indian copyright law. The arguably required overhaul of India’s copyright exceptions will have to address similar policy questions that are currently being debated in the UK.

Substantively, questions that are equally debated in the EU and the US have surfaced against a much more rudimentary statutory background: whether commercial uses of protected subject matter require authorization. Here, the Amici are in stark disagreement, which also seems to reflect the respective normative preferences of the Amici.

While Arul Scaria’s arguments are suggestive of how the law should be read, i.e. equating the machine learning with human learning in the light of the broader implications of AI on the Indian economic and innovation ecosystem, the arguments advanced by Adarsh Ramanujan seem to highlight the current position of law i.e. what the law is and how the acts of Open AI are infringing copyright unless it is demonstrated that they are exempted under Section 52.

Ramanujan’s approach aligns with the written response submitted in the Upper House of Parliament in 2024 by the Union Minister of State for Commerce and Industry (subsequently published by the Press Information Bureau), which stated that the existing legislation obligates the user of generative AI to obtain permission to use the copyrighted works from the owner of such work if the use was intended for commercial purposes i.e. if the use of such copyrighted work was not exempted under Section 52 of the Act.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

This week, COMMUNIA released a new report detailing unfair practices in the licensing of digital resources to libraries (as a PDF file). This report describes contractual practices identified by licensing managers from public and academic libraries across Europe during a meeting organised by COMMUNIA under the Chatham House Rule. The report also contains clauses from licensing agreements sourced from academic library consortia and individual libraries in Europe ahead of the meeting. Together they provide evidence that freedom of contract is often exploited by publishers to the detriment of libraries and access to knowledge.

The report details a range of obstacles arising from unfavourable licensing practices: overly restrictive terms of use that limit scientific research in the digital ecosystem, including text and data mining (TDM) and artificial intelligence (AI) prohibitions; overly broad liability and non-compliance clauses; and obstacles to building and lending digital collections from refusal to license to licensing models that are not adequate to the needs and specificities of the institutions.

Contractual practices that restrict scientific research

A persistent contractual practice limiting scientific research are restrictive AI clauses: participants explained that there has been a noticeable push by publishers to prohibit library users from using e-books and e-journals in combination with AI systems since 2023. Participants reported that, in addition to prohibiting AI uses, publishers have become emboldened and have begun to insist on clauses prohibiting TDM. These trends were also noted in a forthcoming study of 100 licenses by Ana Lazarova.

A general ban on the use of AI tools prevents common uses by researchers and library patrons, such as the use of AI-powered browsers, as exemplified by this clause:

The use of automated IT applications, in particular AI applications, which analyse the technical properties of the database and/or retrieve, analyse, modify, enrich, aggregate, use to generate texts or other content (…) is not permitted. In particular, the use of browsers with integrated or connected functionality of an automated IT application, especially an AI application, is also prohibited. (…)

Many of the uses restricted by the analysed clauses are protected under the national TDM exceptions that implement Article 3 of the DSM Directive. However, these contractual restrictions create legal uncertainty for library users. Furthermore, when the license agreement is subject to a foreign law, the library and its users may not be able to assert their rights under such exceptions. Even though those exceptions are generally protected against contractual overrides, as mandated by Article 7(1) of the DSM Directive, a choice-of-law agreement may be enough to circumvent those exceptions, if none of the protective mechanisms under the Rome I Regulation applies.

Contractual practices that penalize libraries for actions of users

When such contractual prohibitions are in place, publishers can simply remove the library’s access to the licensed content, if they suspect that a user is trying to conduct TDM on their content or using it in connection with AI tools. One licensing manager explained that, as access is provided via the publisher’s platform, publishers can, and have in the past, cut the entire campus’ access to the publisher’s site as a consequence of a single user’s alleged mis-use of the licensed material.

The breach of the TDM and AI restrictions can expose libraries to strong penalties for non-compliance, as exemplified by this clause:

In the event that Publisher determines, at its sole discretion, that Licensee has breached the terms of this section, Publisher reserves the right to take any actions it deems necessary, including, without limitation: (a) immediate termination of this Agreement, without refund or additional compensation to Licensee; (b) denial of Licensee’s existing or future access to any Content; and/or (c) legal action, including but not limited to, injunctive relief, monetary damages, civil or criminal penalties, and/or other available legal remedies.

The sense of vulnerability among the practitioners from the library sector is increasing also due to the fact that there has been strong pressure from publishers to broaden the liability of libraries. Participants described that institutional liability has always been limited as long as the institution takes reasonable measures to prevent license infringement, such as requiring user authentication and informing users of the license terms. However, recent license offers expose libraries to unprecedented liability standards for actions of library users.

As a result, many licence negotiations in the past year have reached a deadlock due to disagreements over the liability clauses. A publicly known example of this are the negotiations between the FinELib consortium and the American Chemical Society (ACS) for the 2025 consortium agreement, which are reported to have concluded without success due to “fundamental differences of views on the liability issues.”

Contractual practices that affect the building and lending of digital collections

In addition, the report also shows how the near total reliance on licenses to access digital resources gives publishers tremendous power to refuse to deal with libraries and to segment access through inadequate licensing models and arbitrary withdrawal of titles.

The licensing models on offer to libraries have been previously documented and include bundling individual titles in a single (often expensive) licensing package, leading academic libraries to spend library budgets only on a handful of publishers and on titles they do not need; metered licenses, which behave as if there was a “self-destruct” attached, forcing public libraries to repurchase licences for the same titles on a more regular basis than they would buy a physical book; and one-book-per-student licenses or “named” licenses, which restrict access to unique users, which is unaffordable for most academic libraries and raises privacy concerns, due to the potential monitoring of student reading behaviour.

Publishers reserve the right to remove or replace licensed content at any time and often without notice. While removing access to titles may be justifiable in some cases (e.g. legal grounds), an arbitrary withdrawal (e.g. retracting course-relevant materials just before the start of the university term or removing titles from public libraries after the selection of the title for an annual reading promotion) is more difficult to accept.

Conclusions and recommendations

Libraries face significant obstacles to acquire digital resources under fair and reasonable terms. The near total reliance on licenses, which are subject to the principle of freedom of contract, combined with the monopoly position of rightholders, enables publishers to unilaterally determine the access models on offer to libraries, impose conditions on the use of materials that exceed their legitimate entitlement and, in some cases, to refuse to deal with libraries at all.

Two measures would go a long way in remedying this situation and offsetting the economic effect of unequal bargaining power of libraries vis-a-vis publishers: a positive obligation to license or to sell digital works to libraries under fair and reasonable terms (see our Policy Paper #21 on the right to license and own digital materials) and the introduction of private international law protections to ensure that research policies are not circumvented by choice-of-law agreements (e.g. qualifying the scientific research TDM exception as an overriding mandatory provision within the meaning of Article 9 of the Rome I Regulation, to set aside the foreign law chosen if it leads to a loss of research rights).^[1]

Resolving these issues would ideally be part of a comprehensive regulation, a Digital Knowledge Act, addressing more broadly the needs of libraries and other knowledge institutions in the digital environment.

This post was first published on the COMMUNIA blog.

———————————————————————————————————————————-

^[1]A recent report prepared for the European Commission explores this and other mechanisms to protect authors, performers and producers from the effects of choice-of-law agreements included in rights transfer agreements.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

This two-part blog looks at the provisions that exist in library laws across European countries concerning the building of collections and what libraries can do with them. It then assesses how far the achievement of these mandated functions is frustrated by a lack of access to e-books.

The first part of this blog provided an introduction to the topic of the interaction between culture laws (library legislation addressing collections) and copyright laws (as expressed in the margin left to rightholders to determine whether and how to allow libraries to licence and lend e-books). This highlighted the connection made between the work of libraries to build and give access to collections and the wider right of access to information. It also underlined the emphasis on the independence of practices here, with the primary guide being to meet the needs of the community and provide the widest possible range of information.

This second part looks at additional expectations and mandates for libraries, and interrogates how restrictions on e-book access risk undermining these. As a reminder, full extracts from laws, with links to original texts, are available here.

Libraries should collect diverse media

While many laws simply do not specify any particular type of media, some explicitly make clear that libraries should be collecting materials in different forms. Flanders calls for a ‘pluriform’ collection, while Bremen, Greece, Trento, Venezia-Giulia and Moldova talk broadly about a variety of media.

Croatia, France, Nord-Rhein Westfalen, Moldova and Slovenia indeed specifically talk about the need to collect and give access to digital materials. The Netherlands does the same, although gives the National Library a specific role in building collections.

The importance of lending

The role of lending in libraries’ ability to fulfil their missions is also set out in a number of laws. Flanders’ law underlines that a duty of libraries is both to offer consultation of a wide offer of books, and:

The lending of materials and files with the lowest possible barriers, in particular for hard-to-reach groups and people with limited incomes.

Estonia’s law sets out:

Loan for in-house use and home lending of items and granting access to public information through the public data communication network are the basic services of public libraries

Similarly, the English, Finnish, Irish, Spanish and Swedish laws also highlight the fundamental nature of lending.

Libraries are also expected to share works between each other, in order to meet needs. The importance of inter-library loan is set out in the roles given to libraries in Czechia, Finland, Poland and Slovenia.

An interesting specific angle is the call in the laws of Moldova and Spain to enable the digitisation of physical materials and then access to these. This makes no reference to being limited only to public domain works.

Building and preserving collections 

Another role of libraries – already hinted at earlier – is to build permanent collections and preserve these for the future. The Danish law, for example, calls for libraries to collect ‘an appropriate proportion of published Danish works’, while the Northern Irish law also makes a connection with safeguarding local heritage.

The Estonian law also underlines that collection building is part of the definition of regional and coordinating libraries. The same goes for Lazio, Basilicato, Emilia Romagna, Piedmont and Poland.

While this is again a role more typically associated with national, research or more specialised libraries, some laws stress the place of public libraries in preserving content, especially as concerns materials about the local area. France’s law, similarly, sets out the role of public libraries in preserving works, and so supporting research, as do those of Poland and Lazio amongst others.

The incompatibility of restrictions on e-book access and lending with library laws

As highlighted in the introduction to the first part of this blog, copyright law has not kept up with the practices of libraries in a digital world. What could have been a major turning point – the decision of the Court of Justice of the European Union in VOB vs Stichting Leenrecht – has been undermined by its lack of reference to the possibility of disapplying contract terms or circumventing technological protection measures, as well as the slowness of governments in implementing this aspect of EU law. None have done so fully, with the Netherlands drawing on the judgment to come to a solution which still allows publishers to refuse licenses (a similar situation to that in the United Kingdom).

Crucially, there is also nothing to prevent rightholders themselves from making the most of the powers they have, and simply refusing to licence e-books to libraries. While Geiger and Jütte make a strong case for an access right, it remains the case that libraries either struggle to gain access to works in the first place, or can see this access withdrawn from one day to the next with little if any notice. In short, European copyright law as it stands today very much leaves access to e-books up to the discretion of rightholders.

Yet, across the examples of laws provided above, this discretionary power is not cited as a relevant factor for libraries when determining (digital) collections policies. Indeed, these purely commercial decisions may indeed contrast with the call for libraries to be free of commercial pressures in some laws. While of course libraries need to work within the limits of their budgets, further restrictions based on the discretion of rightholders seem incompatible with true library independence.

In addition it is clear that when a more or less significant share of e-books are unavailable to libraries, it is far harder to meet the conditions set out. Building universal, balanced, responsive, up-to-date collections is far harder when working from a more limited catalogue of works.

With the act of lending itself set out as a core function of libraries too, contract terms and technological protection measures that make it impossible to lend – both to individual users, and between libraries – are also problematic, confounding the goals established by legislators.

Finally, the model of e-book lending most commonly on offer, with e-books held on publishers’ or other platforms’ servers rather than acquired by libraries, also stands in contrast with those library laws that urge libraries to build collections, as well as to practise preservation. Enough laws are clear that e-books have as much of a place in library collections as physical ones.

In sum, it would appear that an ongoing failure to protect the ability of libraries to acquire, lend and preserve e-books runs directly against the will expressed by legislators when passing library legislation.

Conclusions

This article has set out to explore how library laws from 33 jurisdictions in 17 countries across Europe discuss the building of library collections. It looks in particular to identify common trends between them, notably regarding their integral importance for access to information, and the need to respond to local needs while also offering as wide a range of perspectives as possible. Other shared features include an openness to a variety of media types – with digital explicitly named in some – and then the possibilities to lend and preserve collections.

The article then underlines how restrictions on libraries’ access to e-books, and possibilities subsequently to lend or preserve these, serve to undermine the achievements of these goals, set out by governments and legislators as part of their broader cultural policies.

The impact of this, in addition to the frustration of cultural policy goals, is also wider. Library laws – and in particular provisions on collection building – frequently are designed as part of wider efforts to promote reading, literacy and education (France, Schleswig Holstein, Rheinland-Pfalz, the Netherlands, Sweden). Some go further, noting how this work supports social inclusion (Puglia), democracy (Nord-Rhein Westfalen) and development in general (Estonia, Finland, Rheinland-Pfalz). Insofar as collection building and lending is hindered, so too is work towards these goals.

In short, the ongoing lack of steps to ensure that libraries can access and lend e-books on similar terms to physical ones represents a continuing drag on the achievement of cultural laws and wider public policy goals.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

This two-part blog looks at the provisions that exist in library laws across European countries concerning the building of collections and what libraries can do with them. It then assesses how far the achievement of these mandated functions is frustrated by a lack of access to eBooks. This first part introduces the issue and covers the broad principles that apply to collection building.

The question of the terms under which libraries are able to acquire or access and then lend e-books is arguably illustrative of the situation facing these institutions more broadly in the digital world.

For well over a decade, libraries have sought to develop their offer of e-books, driven both by demand from users and the understanding that they represent an opportunity to offer books in new ways and to new populations.

In the university sector, the shift has been dramatic, with digital materials (both journals and monographs) coming to dominate. While the change has not been so far-reaching in the public library sector, e-books have come to represent an important part of many libraries’ offer to users.

This is not to say that this process has been easy or painless. The copyright laws that enable libraries to lend in the physical world (either through explicit provision for this, or by allowing lending to take place under the exhaustion doctrine) have not kept up with the shift to digital.

The Court of Justice of the European Union in VOB vs Stichting Leenrecht offered an incomplete solution. While it established equivalence in the treatment of physical and digital works for the purpose of public lending, it failed to allow for the circumvention of technological protection measures or the disapplication of contract terms, for example those preventing lending or inter-library document supply, as well as the taking of preservation copies or other standard library activities.

The result is that the ability of libraries to acquire (or access) and lend e-books is largely dependent on the willingness of rightholders (typically publishers) to let them. Over and above questions about the terms imposed on libraries, there is the more fundamental question of whether e-books are available at all to libraries.

There is already an argument under copyright law that there is an implicit right of access. This is derived from the logic that other activities foreseen under law (such as preservation, education and research uses, and lending) can only happen if there is access in the first place. And of course, there is also precedent for addressing contract override, not least through Article 7 of the Directive on Copyright in the Digital Single Market, even if this does not apply to eLending (yet).

However, we do not only need to look in copyright law for arguments in favour of access. This article looks rather at 36 library laws across 20 different countries in Europe in order to explore what they say about how libraries can build collections. It then assesses how these provisions fit with copyright regimes that leave choices around collection building to rightholders.

The article is organised around the different ways in which library laws talk about collection building. This echoes the call in documents such as the International Federation of Library Association’s 1999 Statement on Intellectual Freedom that:

Libraries shall ensure that the selection and availability of library materials and services is governed by professional considerations and not by political, moral and religious views.

A full list of the laws consulted, with relevant excerpts and links, is available here.

Common themes relating to collections policies across laws analysed

Library collections have a special role in delivering on the right of access to information 

A number of laws, in particular those at the regional level in Germany, place the work of libraries in the context of the wider right of access to information that German citizens enjoy in line with the Federal Constitution. This provides a very strong legal grounding for the work of libraries in being able to acquire and give access to works. In this vein, the law in Nord-Rhein Westfalen therefore notes that:

[Libraries] contribute in a special way to the realisation of the fundamental right under Article 5 paragraph 1 sentence 1 of the Basic Law to be able to obtain information from generally accessible sources without hindrance.

There are similar provisions in Thuringen and Rheinland-Pfalz. Meanwhile, the Estonian law sets out that:

The purpose of public libraries is to ensure free and unrestricted access to information, knowledge, achievements of human thought and culture for inhabitants

The Spanish law states:

Public libraries are the means by which the public authorities make possible the effective exercise of the right of all citizens of Access to Information, Education and culture in the context of the Information and Knowledge Society.

A number of Italian laws also connect the work of libraries to the right of access to information.

Library collections should be responsive to the needs of the public

A common theme across countries is that a central driver of collections policies for public libraries is the need to respond to the needs of local communities. There is a duty on libraries to ensure the relevance of their offer. This is the case for Belgium (both Flanders and Wallonia), England, Estonia, Greece, Moldova, Northern Ireland, and Norway. Slovenia and Denmark simply talk about an appropriate range of material.

As an example, Wallonia states that collections should be:

Quantitatively and qualitatively meaningful for the population served, and representative of contemporary socio-cultural needs inherent to the public character of the institution. 

The Greek regulation on the operation of public libraries sets out:

The identification, analysis, and evaluation of the audience’s informational needs form the foundation of the Library’s collection development policy. This identification is based on public research methods (such as questionnaires, usage statistics of materials, etc.). It is essential to note that the fundamental principle of this process is that “all user groups, regardless of age, gender, race, religion, nationality, language, or educational background, must be able to find material relevant to their needs and interests.”

In the case of Puglia, the adaptation of laws to needs is underlined as being key to realising the potential of libraries as drivers of social inclusion. This becomes possible when it is ensured that all members of a community can find books relevant for them, and so all are able to benefit from taxpayer-funded services such as libraries:

Inter alia […], the region promotes actions and agreements directed towards social and multicultural integration and the development of library and documentary collections and of library services focused on the needs of disadvantaged user groups.

In Saxony Anhalt, the specific role of libraries in building collections that meet the needs of employment seekers is highlighted.

The ability of libraries to build a relevant collection or not can also have financial implications. For example, Wallonia underlines that libraries can only receive support for engagement in reading promotion activities in line with the quality and relevance of their collections. The level of support depends, inter alia:

On the offer of documentary and cultural resources, in terms of the […] adaptation [of the collection] to the population and the five-year development plan. 

A connected point is that around the need for regular renewal of collections in order to keep them up to date. This is highlighted in the laws of Wallonia, Czecha, Finland, and Norway.

Libraries should be independent in building their collections 

Some laws, echoing the IFLA Statement, stress that libraries need to be independent in their decision-making. In practical terms, this means that they should be able to deploy collections budgets in line with professional principles.

The laws in Flanders all underline the need for independence, including from commercial interests, as do those in France, Schleswig Holstein, Hesse, Nord-Rhein Westfalen, and Rheinland-Pfaltz.

For example, the law in Rheinland-Pfalz includes a section on the independence of media selection:

2. Independence in Media Selection: Libraries intended for public use are independent in the selection of their media content.

In France, the library law underlines that:

[Libraries] must be exempt from all forms of ideological, political or religious censorship, or commercial pressures.

Latvia’s law sets out:

Printed materials, electronic publications, manuscripts and other documents in library collections, regardless of the political, ideological, religious or other orientation of their authors or the information therein, shall be accessible to any person in accordance with the procedures established by the libraries.

Meanwhile, in Croatia, the law simply underlines that collections and holdings should be built up in line with generally accepted professional criteria.

Libraries should promote pluralism, diversity and quality in their collection building 

While this is a goal most often associated with national or major university libraries, some laws also expect public libraries – individually or collectively – to aim for a universal collection, providing information on all possible topics. This is the case in Czechia, Estonia, and Spain. In France, for example, the law sets out:

The collections of public libraries and their networks are pluralistic and diversified. They represent, at their level or in their area of specialisation, the multiplicity of knowledge, currents of thought and opinion, and editorial production. 

Linked to this, the law in Rheinland-Pfalz suggests even that libraries, through their work, should contribute to the knowledge commons:

Libraries are service providers in the modern knowledge society, which understands knowledge as a common good in which every member of society can participate and contribute.

Elsewhere, the goal is simply to build the widest possible collection (as in Trento), or to have one that is versatile (as in Sweden). Sweden’s law also puts a strong emphasis on ‘quality’ as a goal, with the explanation of the law underlining that:

it is, for example, not possible for a municipality to decide that the libraries in the municipality should have a distinctly limited or lopsided range of media.

In summary, library laws are clear that their work to build and give access to collections is closely aligned with the (sometimes constitutional) right of access to information. In delivering on this, the importance of independence, responsiveness to needs and diversity is clear.

The second part of the blog will explore additional principles that apply, and how these relate to restrictions on access to e-books.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135

Regular readers of the Kluwer Copyright Blog may already be familiar with the excellent reviews of the first two rulings on the European Union’s new text and data mining (TDM) exception – one from Germany (see the Kneschke v. LAION ruling here, here and here) and one from the Netherlands (see the DPG Media v. HowardsHome ruling here). The third TDM ruling originates from Hungary. The judgment, issued on 3 December 2024, deserves close attention, as the Municipal Court of Appeals of Hungary had to determine – among other things – whether the scraping of the plaintiff’s website by the leading global search engine, for the purposes of indexing relevant content and providing snippet views, falls within the general-purpose TDM exception under Article 4 of the CDSM Directive.

Importantly, the case was not solely about TDM. The defendant presented five distinct arguments to exempt its activities from liability. First, it argued that the European Court of Justice’s ‘hyperlinking’ case law – especially the ‘new public theory’ – applies to the indexing of press publications. Second, the defendant claimed its use fell within the exception to the press publishers’ ancillary right [Article 15(1), last sentence, CDSM Directive, transposed into Article 82/C point (b) of the Hungarian Copyright Act (HCA)]. Third, it contended that the plaintiff’s reproduction right is merely ancillary and cannot be infringed without an act of making content available to the public – something that was claimed not to occur in this case, as the defendant’s actions remained within the limits of the exceptions under Article 82/C HCA. Fourth, the defendant argued that either the temporary act of reproduction exception or the TDM exception should cover its scraping and indexing practices. Finally, it claimed that the plaintiff implicitly consented to its practices by not excluding the defendant’s bots via the robots.txt file on the plaintiff’s website.

The court of first instance was divided on the relevance of these five defenses. However, the Municipal Court of Appeals ultimately accepted the most significant one: the second argument. It concluded that the displayed snippets fell within the exception to the press publishers’ right. More importantly for the purposes of this post, the Court of Appeals also overturned the trial court’s rejection of the TDM-based defense and its ruling on scraping. (To be clear, the Court of Appeals did not accept the defendant’s arguments regarding the new public theory, the ancillary nature of the reproduction right, or the defense based on temporary acts of reproduction.)

In paragraph 51 of the ruling, the court concluded:

“[i]t was not disputed that the defendant had lawful access to the plaintiff’s press publications and did not circumvent the technical measures taken by the plaintiff to access them. Moreover, the robots exclusion protocol assigned to the plaintiff’s website allowed all search robots, including the defendant’s search robot, to crawl and index, and the plaintiff did not object to the indexing in the form required by law. The robots exclusion protocol is not of significance for consent, but primarily for the fact that it is—in accordance with the provisions of the Copyright Act—a machine-readable form by which the right holder could object to the text- and data-mining by the search engine. Since the defendant does not save the download of the page required for crawling, as it does not keep a copy of the page, and the data and information contained in the index are not copies, the statutory condition in Article 35/A(1)(c) [of the Copyright Act] is also satisfied.” (Municipal Court of Appeals, Case 9.Pf.20.353/2024/6-II, 3 December 2024, para. [51]. The Hungarian-language decision may be found on the judiciary’s website using the keyword 20.353/2024.)

In short, the ruling declared that web scraping and search engine indexing constitute ‘a form’ of TDM. This conflation of scraping with TDM is not unprecedented. Measures I.2.2, I.2.3. and I.2.4. of the Third Draft of the General Purpose AI Code of Practice (see more here) take a similar position.

This interpretation appears to reflect the interests of the platform industry, however, and it has also sparked significant criticism (see here). This post takes the position that such a conclusion does not align with the overarching purpose and substance of the CDSM Directive. Accordingly, I feel compelled to critique both the Hungarian judgment and the Third Draft of the Code of Practice.

First, although the CDSM Directive defines TDM in a broad, technical sense (Article 2(6)), its telos was never to exempt all forms of automated data analysis from liability. The recitals of the directive make this clear: the “processing of large amounts of information with a view to gaining new knowledge and discovering new trends possible” may be carried out for research purposes under Article 3 of the CDSM Directive, or for “government services, complex business decisions and the development of new applications or technologies” under Article 4 (see recitals 8 and 18). None of these objectives appear to encompass the decades-old practice of web scraping by search engines.

Recital 9 of the CDSM Directive also explicitly states that “there can also be instances of text and data mining that do not involve acts of reproduction or where the reproductions made fall under the mandatory exception for temporary acts of reproduction provided for in Article 5(1) of Directive 2001/29/EC,” in which case Article 4 of the CDSM Directive is not applicable. Moreover, Article 4(2) specifies that any reproduced or extracted information may only be retained for as long as necessary to carry out TDM—where the emphasis is clearly on the mining, not on any subsequent services the user might provide.

The ruling of the Municipal Court of Appeals, however, appears to legitimize the permanent storage of collected information for purposes beyond mere mining – namely, for indexing. This interpretation extends the scope of the general-purpose TDM exception in a way that seems unjustified.

Second, including web scraping under Article 4 risks violating the three-step test. The court’s ruling conflates distinct technological processes and subjects them to a single legal provision. Rather than clarifying the law, this approach introduces legal uncertainty and increases the complexity of interpreting and applying the CDSM Directive.

The general-purpose TDM exception – outlined in Recital 18 – was crafted to comply with the three-step test. This is evident from the built-in safeguards of the directive: the lawful access requirement; the purpose-limited nature of the reproduction or extraction; the restriction on retaining collected data only as long as necessary for mining; and the right reservation under Article 4(3). (Of course, differences in implementation may exist across Member States. For example, the Czech transposition of Article 4 into Article 87b of the Czech Copyright Act does not explicitly include the ‘lawful access’ condition.)

However, applying Article 4 to web scraping – arguably the most essential underlying technology of internet browsing – extends the exception far beyond “certain special cases” as required by the first prong of the three-step test. Such an interpretation effectively transforms the exception into a general rule.

Furthermore, the third prong of the test – prohibiting “unreasonable prejudice to the legitimate interests of the rightsholders” – also weighs against the court’s interpretation, for at least two reasons.

First, while rightsholders may allow indexing of their content to facilitate discoverability, this does not imply consent for broader commercial uses of their content. Although robots.txt can (albeit crudely) distinguish between basic scraping and more advanced forms of mining (see Hanjo Hamann’s paper on this point), the defendant challenged only the plaintiff’s failure to block scraping—not TDM—via robots.txt. The refusal by both the defendant and the court to differentiate between indexing and automated content analysis undermines the ability of rightsholders to make informed, nuanced decisions about the use of their content by different technologies. If scraping for purposes other than automated analysis were subsumed under the general-purpose TDM exception as well, rightsholders would face the unfair situation, where – either due to the lack of technical expertise or because of the ‘fear of missing out’ from indexing – they would tend not to exclude scraping of their websites. Also, as a consequence to it, they could effectively lose their expressly granted statutory right under Article 4(3) of the CDSM Directive to opt out of TDM – and, with it, their ability to authorize such uses.

Second, indexing has long been recognized as a core function of search engines and is already protected under the E-commerce Directive’s safe harbour provisions (currently under the Digital Services Act). It is, therefore, not a practice generally subject to legal challenge. But this exemption should not be extended to the domain of TDM.

The author’s manuscript on rights reservation, the AI Act, and the evolving TDM case law is available via SSRN here. See further Martin Senftleben’s comments on the right reservation prong of the TDM exception on the Kluwer Copyright Blog here.

More from our authors:

	The EU Artificial Intelligence (AI) Act: A Commentary by Ceyhun Necati Pehlivan, Nikolaus Forgó, & Peggy Valcke € 285
	Originality Standard of Photographic Works in EU Copyright Law by Marián Jankovic € 105
	Regulatory Challenges of AI Governance in the Era of ChatGPT by Md. Toriqul Islam € 135