The European Writers’ Council (EWC) expresses its sincere gratitude for the important and long-awaited strategy and recommendations outlined in the own-initiative draft report Copyright and generative artificial intelligence – opportunities and challenges (2025/2058(INI)). 

Brussels, 14th July 2025 : As the EWC is the world’s largest representative body for only writers in the book sector – with 53 writers’ organisations from 34 countries of the EU, the EEA and of non-EU areas, representing 250,000 writers and translators – we are taking the opportunity to comment on the draft published by Axel Voss on 4 July in the version dated 27 June 2025 for the members of the JURI Committee. 

We are aware that the report will evolve until the planned committee vote and would like to provide you with expert feedback from the authors’ perspective as the genuine rights owners of all new hitherto rights related to AI and generative AI development, throughout the process. Accordingly, this initial and early response by that group of cultural innovators whose works (and work) are massively impacted by the approximately 270 large language models with 10^23 FLOPs and 1328 LLMs in operation worldwide, should be perceived as the beginning of what we hope will be a valuable exchange. 

Overview of the feedback 

• Preamble 
• On the draft’s Context 
• On the draft’s Recommendations – and some fair warnings on the proposed solutions 
• A plea for precise vocabulary and evidence-based arguments
  
  Download the statement (pdf, 7 pages)

Preamble 

We are convinced that authors’ rights, personal rights and anti-discrimination laws have by no means lost their relevance – on the contrary, they are more important than ever. 

–> What is lacking is a decisive stance and effective instruments to enforce these rights. 

On the draft’s Context

I Authors’ rights, copyright and human rights are by no means an ‘obstacle’ to technical development (9, L), a fact which Prof. Nicola Lucchi also states several times in his recently published study “Generative AI and Copyright”, commissioned by the JURI Committee.

Technical advance often is a collaborative activity between science and the arts. The claim that copyright is restrictive on technology has a prosaic rather than evidence-based foundation: it arises from the unwillingness of companies to pay appropriately – or, in the case of generative AI, which generates competing products and massively restricts us writers in the exercise of our rights and interests, violating the Berne Convention, European legislations and, in particular, the three-step test – to pay substantially.

• It is not the legal situation that is ‘impeding’ European technology markets (B), after all, copyright in the form of patents incentivises EU entrepreneurs who are confident their innovations will be protected by law: it is an aversion to pay for works and labour behind each creative work.

  

II Reproducing and imitating technologies, known as ‘generative AI’, must be considered with precise thresholds:

First, there must be clear definitions of which systems are meant, for which use, and in which market they are being used (or rather, whether they operate in the free market or in exclusive-use environments).

Secondly, the input and output segments must be clearly defined, especially when it comes to the question of payment:

• Compensation, i.e. damages, will become increasingly relevant in relation to output: market damage, reputational damage, fraud involving public funds, loss of tax revenue, declining social security contributions, damage to each member state that goes far beyond the primary value chain.
• This is distinct from remuneration, i.e. payment for the use of a work for the input into the development of systems, and in the context of AI development in particular: the use of the labour time and quality of the human effort involved, not just the use of a finished work (as e.g. in current licensing systems).

 

III  Usage agreements could have been in place long ago if AI developers had in mind to benefit society, to respect legal usages and ever seriously considered licensing creative works for training purposes.

Global rights management societies (CMOs) can offer national repertoires worldwide or operate through reciprocal agreements. Legislation would even have been suitable for this before 2019 – it has been proven that AI developers have been using protected works since several decades without obtaining authorisation or payments, and most often from dubious sources including piracy websites.

Fact is, that the majority of book authors – as the genuine owners of (gen)AI rights and of the material used by AI – remain strictly opposed to the use of their work and their works for the development of generative technology in general. They do not want to support a system developed to replace them and other jobs. Our recent EWC surveys, which we will share with you upon request, clearly show that, while our members do not want to prevent licensing, over 50% claim the right to refuse any use under any condition. Those few (ca 15%), who are ready to license, want full control, transparency, limited usages, and a substantial payment.

• Accordingly, we reject any mandatory licenses including mandatory collective licensing, and also strongly reject the expanding of Art 4(3) 2019/790 (EU) Directive onto GenAI use (7),), or any addition of a new ‘AI exemption’ (explanatory statement) which again would make use of an opt-out-regime. Instead, opt-in, as usual for exclusive and non-waivable rights, shall be promoted.
• Authors must be granted the right to exercise their exclusive rights, not through a regime of post-facto reservation of rights (“opt-out”), forcing authors into illegitimate formalities, but exclusively through the voluntary granting of usage agreements for GenAI uses and with substantial
• Also please note that most authors have not transferred contractually any TDM or (gen)AI rights to contractual counterparts. All your considerations must take therefore the single author into account, and all measures must be feasible for an individual author as the first impacted rights owner, and to be handled independently from their publishing partners or distributors, including exclusive rights management and / or licensing.

IV   The European Parliament and its most important statutory body, the JURI Committee, have a duty to defend authors’ rights in this time of upheaval and negotiation between the economic ambitions of tech giants and the principles of democratic societies with their rights to culture, media and freedom of speech.

This power comes with responsibility: to protect rights, and to defend cultural plurality.

It is of paramount importance that the committee does not undermine the backbone of authors’ rights law and does not place more and more burden on single authors to defend their rights, such as costly technical standards like the long-outdated and removable ‘watermarking’, or only one way of machine-readable rights reservation. Most importantly, there must be no introduction of protection that is solely dependent upon the rights holder leaving a record in a register (9; explanatory statement).

• The author’s right must not be dependent on appearing on a list or using one only (N) costly technology for declaring a rights reservation – authors’ rights as protection of property, work, decision making on the own work, have undergone centuries of progress and negotiation. It should not be destroyed by compulsory registration, or by restriction of diverse and sector specific expressions of opt-out, as all these measures would render it meaningless. It would render the Berne Convention meaningless, too.
• It is of utmost importance that those who want our high-quality work also make a significant effort to enable this high-quality work and creativity to happen, i.e. by offering licensing requests and establish frameworks of documentation and negotiation. AI providers are called to incentivise human creative work and human labour. It should no longer be the norm that nothing is expected of wealthy AI providers, which systems are entirely dependent on authors’ work, but everything is expected of the creative sectors and their vulnerable individual actors. 

V  Legal debates have changed significantly to the detriment of established rights.

Whereas introducing Exceptions & Limitations (E&L) used to be about serving the public – for accessibility, for supporting institutions for the blind or for educational purposes – since the rapid digital evolution, it has increasingly been about business interests alone. Above all, for monopolies based outside the EU and protecting their millions in profits, while pushing Europe into a digital dependence from their overall-all presence. The value from generative LLMs/text, image or video generators to society is not proven: indeed, those reproducing and imitating technologies had provided very little gain for humans. We monitor increasing job and skill losses, the collapse of markets, devastating resource consumption, intergenerational inequity, manipulation, political propaganda and misusing of generative AI to produce fake news, fake porn abusing even the images of minors, and fake events.

• This is no longer about striking a balance between the legitimate objectives of a society. The balance has long since been lost, and any further E&L to serve AI is at the expense of vulnerable individual authors.

 

On the Draft’s Recommendations, and Some Fair Warnings by the EWC

 

1. On Art 4. 2019/790: 6, J, V – a missing link – and a new “AI” exception: item 7

We greatly welcome the clarity of the report, which finally casts doubt on the scope and over-interpretation of Article 4 of Directive (EU) 2019/790 on Copyright in the Digital Single Market (CDSM Directive) as EWC has been sustaining for years.

Back in 2019, during the Trilogue negotiations on the CDSM Directive, the EWC warned MEPs in direct exchange that technology companies would abuse this exception to develop software that will compete with those authors whose works and labour will be exploited unpaid and non-authorised.

MEPs and national MPs tried to reassure us at the time that this was not the intention.

 Six years later, we find ourselves in exactly this foreseeable devastating situation where the damage caused by the unregulated deployment of generative software is nearly irreparable: markets have been eroded; writers, translators and illustrators have lost contracts and workplaces[1] and are paying less in taxes and pension contributions. Unlabelled AI products that falsely claim to be ‘books’ obtain remuneration from collective societies, are unlawfully subsidised by state-financed funds through publishing or translation grants or undermine the income of authors’ due to offering the products with dumping prices that are not reflecting the true cost of AI products in regard of their consumption of water and energy as well as causing emissions.

• Accordingly, the move to question the exception of Art. 4 CDSM (6) is long overdue, and the damage done so far must be assessed, as outlined in the explanatory statement. All damages done from – mostly non-EU-oligopolies – should be assessed in parallel to the draft INI report, and as laid down in item 10 of the Parliament resolution on Intellectual property rights for the development of artificial intelligence technologies of 20 October 2020.

 

The missing link: Article 3 of the Directive 2019/790 (EU)

However, we do not see the same criticism being levelled at Article 3 of the CDSM Directive. If, even in light of Nicola Lucchi’s study, AI development and its various stages, such as long-term storage, reproduction, memorisation, making available to the public, etc., are not covered by text and data mining, why should scientific authors, for example, continue to suffer from the exploitation of their work by commercial AI providers? Article 3 is similarly abused by commercial AI developers: through private partnerships between scientific and research institutions and AI developing companies, entire sets of works collected under Article 3 are transferred to commercial use after the research has been completed and are reused for commercial AI purposes without the authors being granted any rights reservation. Generally, we hope that the report will clarify Article 3 to the effect that further (generative) AI development, after TDM in research, is not comprehensively covered by Article 3.

• This is particularly relevant as the Code of Practice for the AI Act does not address this most common use, including the disclosure of the provenance of third-party data sets.

However, we strongly caution against pursuing the path outlined in point 7, which would result in a new exception with an again predictably dysfunctional reservation of rights or extending the existing TDM exception to (gen)AI development.

• We reject any additional (Gen)AI exception in full.

The development of generative “AI” systems such as LLM and GPT, and the various stages of AI production are not equivalent to text and data mining, both from a legal and technical view. Studies by Nicola Lucchi, as well as Prof. Dornis and Stober, confirm this sufficiently.

• TDM is not AI development. Expanding the existing exception shall not be

Enlarged European legislation should not allow continued injustices towards its own authors: as the report itself confirms, the forms of use by (Gen)AI developers are a new form of use of creative works and, accordingly, require compliance with the originators’ rights. These remain, whether under EU authors rights and copyright law or the Berne Convention, exclusively with the owners: the authors. It would once again run counter to the spirit of global IP law to reverse the exclusive right and the voluntary ‘yes’ – to use under negotiable conditions – by means of a subsequent reservation of rights.

• This proposal runs counter to the entire draft report. We deeply regret this destructive approach of an expanded TDM exception or an additional GenAI exception with opt-out regime, as the report otherwise contains numerous excellent and valuable suggestions:

 

2. Transparency, opt-out-directory, central licensing hub via EUIPO

We expressly thank the rapporteur and the drafters for identifying the need for title-specific transparency (10, 11, O, R) as the basis for enforcing legitimate rights and for calling on AI providers to confirm their compliance with the reservation of rights. This fills the gap the Code of Practice is leaving us – with no chance to exercise our rights at all. Indeed, we’ve had until this time, NO chance to exercise our rights. Not at all, other than with costly court cases, all still pending.

We also appreciate the reference to the transparency of AI products (U). As stated in the part about the draft’s Context, these products unlawfully mislead consumers, particularly when they are sold at the same prices as genuine cultural goods, disrupt markets, obtain unlawful remuneration through CMOs or public funds in the form of subsidies, prizes or grants, or conceal their AI origin and harm contractors.

However, we would like to comment on the proposed solution that follows:

For us, a REGISTER that only grants protection upon completion of the sign-up process would not be appropriate. We mentioned this in the context: administrative registration must not replace the basis of a legal protection; every work should enjoy the protection of this reservation of rights, provided that a reservation of rights is affixed in all practicable and readable forms.

• As book authors, however, we are considering creating a data DIRECTORY of rights preferences. This could be done at the EUIPO with the help of the ISCC (International Standard Content Code) and a rights declaration, or by means of other identifiers in the book sector, such as ISSN, ISBN, DOI, etc. This could also encourage AI developers to recognise the reservation of rights in the long term and enable them to confirm the respect.

Whether license management via a central body makes sense for all sectors equally remains to be very carefully explored before it is included in a report or resolution. Few CMOS manage primary rights worldwide – in some sectors. In others, as in our profession, only secondary rights are managed nationally or with worldwide bilateral agreements. Also, collective licensing is not equally welcomed among all cultural actors. This needs to be respected and no mandatory collective licensing imposed. License negotiations, especially in the field of AI, require a multi-approach that breaks with traditional licensing visions: AI use is a form of use that does not add value to a work. It does not involve the dissemination of names, revenues from sales abroad or a Netflix series. Additionally: labour, experience and working time are exploited. Accordingly, remuneration for use must be determined very carefully according to sector and type of cultural goods; in the book sector, for example, we consider five-figure sums to be the minimum for one-time use of one title for Gen AI.

 

3. On technical aspects: We would ask you to refrain from setting singular technical standards for rights declarations (N) or for tracking works in AI systems via only one measure (Q).

Not all forms and distribution methods of cultural works can adopt ONE standard – just as books cannot use robots.txt, photographers cannot use ISBN, watermarks do not work for performances or audio works. Watermarking is thus far from a sustainable solution: it can be easily removed and, moreover, cannot be implemented in all cultural works.

Example: if a print book is scanned and then used for AI despite an opt-out (see the Anthropic case), no watermarking in the world can help. The same applies to audiovisual or audio formats such as audiobooks: the development and implementation of a ‘watermark’ is not possible. Also, work sets from third parties, e.g. research entities, come as full “work corpora” outside the original format. Any watermarks or asset-based meta-data lost. Only external identifiers, like the International Standard Content Code (ISCC), could support our claims here.

Apart from that, numerous markets and individual book sector players in Europe will hardly be able to afford the technical costs, and it is questionable why rights owners – again – should have to spend hard earned money to protect their rights mechanically, while AI developers do not even bother to check works for rights reservations. This has been seen in the lawsuits against Meta and Open AI: many of the works included in the set used via the piracy website Library Genesis or Anna’s Archive and Book3, had an opt-out implemented. Whether this has been respected and sorted out of the pile, is questionable.

• Conclusion 1: The focus must be on confirmed respect for EVERY opt-out, in every form and adjusted per sector. This is the only way to achieve the level playing field you are striving for: to have a mandatory opt-out confirmation-system, built by AI providers.
• Conclusion 2: There can only be one solution to rectify the TDM exception: any use of (gen)AI does not fall under this exception. The use of (gen)AI is an exclusive right and requires the voluntary, informed and negotiable consent of the author.
• Only this will strengthen a licensing market, and if the ART principles are respected: authorisation, remuneration, t
• A market for quality texts, and with the willingness on the part of authors, cannot be achieved through a further exception.

 

The EWC pleas for precise vocabulary and evidence-based arguments

1. We do not produce “data” or “content” (and an AI system does not ‘create’ or is ‘trained’): please adhere to appropriate vocabulary that is legally consistent with existing provisions, values the labour and the works of authors, and reflects reality.

The draft report is characterised by a high appreciation of the achievements of the cultural and creative sector. However, these achievements are entirely based on the authors, without whom there would be no value chain – and without the authors’ years of labour on a single (book) work, there would not even be one generative technology. We are the main source of this ‘innovation’, and without us, there would be none of these reproducing and imitating systems running. It’s our long years of labour and development on writing which is exploited by (Gen)AI providers to develop their models. All development of AI has to be human centred for the sake of a prospering Europe and society, not forcing humans to further exploit themselves for any type of generative machines and the profits of a few non-EU oligopolies.

Accordingly, we encourage you to remain appreciative in the vocabulary used in the report. We neither produce ‘data’ nor ’content’, and we plea to name our achievements what they are: works, cultural goods, texts. This will help to distinguish clearly between human creation and AI output. In some paragraphs, both is currently referred to by the same term, see for example D, F, H: ‘Content’. AI products should be referred to as products or output, protected works as creative works or human creations.

 

Please be careful not to diminish the achievements of human authors and not to exaggerate or even anthropomorphise the supposed expertise of statistical calculation models (D, H). A linguistic distinction will sharpen the report and give the debate the necessary development and precision.

Generative software calculates or produces. It does not ‘create’ anything; it combines tokenised fragments on the basis of mathematical probability calculations, drawing on memorised partial or complete works. In Prof. Lucchi’s words, GenAI reproduces and imitates. Similarly, the word ‘training’ is commonly used but is technically incorrect: only the number of tokens (themselves created on the basis of books and their translations) and segments is increased, and the parameters are amplified further using computing unit power to have more algorithmic alternatives. ,Training’ in the sense of human education or intellectual development does not take place. Here, too, stick to language that reflects reality: this is about the development of systems and software, not ‘training’.

• A precise wording is necessary not only to avoid marginalising and ‘dehumanising’ the work, investment and professional attitude behind every book, but also to remain consistent with legal texts such as the AI Act, which refers to ‘content sets’ rather than ‘data sets’ – or to avoid provoking legal uncertainties, such as the question: if our works are ‘data’, how does the GDPR come into play?

 

2. There is no valid evidence that IP rights are ‘hindering’ technical development (B, 9, explanatory statement), a fact which Prof. Lucchi also states several times in his recently published study “Generative AI and Copyright”, commissioned by the JURI Committee.

It is not existing IP law that stands in the way of technological development, but rather the unwillingness to allocate budgets and pay for the use of protected works. As the developments in so-called ‘generative AI’ clearly show, the law has not prevented oligopolies from setting up systems. Patent laws are the exact proof of this. The technological development of AI depends much more on research investment in powerful chips, server farms and the skill development of IT engineers.

Furthermore, real technical breakthroughs will be found in general-purpose AI and not in LLMs, which have been stagnating since their initial development in the 1970s.

• The only issue that opposes IP rights is the cost of using existing copyrighted works. However, if technology companies are willing to pay for its use, a solution has long been possible within the existing legal framework – it is purely the lack of willingness to pay for the essential basic ingredient.

Final remarks

We thank you for this strong initiative and also for innovative visions in the explanatory statement such as compensating for damages by a percentage (5-7%) of the AI companies’ turnover – but the value of what we have had to supply due to the misinterpretation of the TDM exception(s) runs into more than millions of euros and will be more as only 7% GenAI turnover. At stake here is the economic value of our human authors’ works from Europe, also contained in large collections such as Library Genesis. This sums up to the hundreds of billions – just from what authors have invested individually in their works, and not included yet the calculation of e.g. scientific authors, exploited under Art. 3 (2019(790). There, public money for state institutions had been invested to create scientific works – and this state investment likewise is misused by non-European enterprises under the cover of Article 3 and private partnerships.

Likewise, all in the draft proposed lump sums will lead to disputes about the distribution, assessing the values of each sort of work included – and can only work with a full transparency on which works had been used when and what for. Plus: all lump-sum payment for the harm done has to not be interpreted as legitimising the very likely not legal exploitation by AI providers. Nevertheless, this call for compensation of the nonconsensual use of our works draws an important line in the sand, a momentum, which we highly welcome.

We kindly ask you to consider our feedback during the negotiations of amendments and remain at your disposal for exchanges and background information from the writers of the book sector.  The EWC AI Task Force is available in several languages to answer your questions or to join in a fruitful exchange.

With our kindest regards:

 

Nicole Pfister Fetz                    Nina George                                           Zsuzsánna Dóczy

Secretary General                       Commissioner for Political Affairs          Coordinator

Maïa Bensimon                       Monika Pfundmeier                           Claus Ankersen

Vice President, France              Vice President, Germany                      Board Member, Denmark

Conor Kostick                        Sebastià Portell

Board Member, Ireland             Board Member, Spain

[1] https://europeanwriterscouncil.eu/soa-survey-uk-ai-2024/