Joint open letter by CEATL, EFJ and EWC to Executive Vice-President Virkkunen & the EU AI Board

Dear Executive Vice-President Virkkunen,

Dear Members of the EU AI Board:

We, the three federations CEATL, EFJ and EWC, represent over 550,000 individual authors from 159 associations in the text sectors, who work as writers, journalists and literary translators in all genres and media forms in the EU and beyond.

We are writing to express our strong opposition to the third draft of the EU’s Code of Practice under the EU’s AI Act legislation. The simplified and industry-friendly orientation of the Code of Practice, combined with a vocabulary that rarely calls for commitment, means that EU law will be undermined, and the minimum requirements of the AI Act will not be met.

‍

Ignoring Feedback

Regretfully the third draft of the Code of Practice continues to ignore the substantial feedback of the authors, as the original rightsholders, despite our active and argumentative involvement in the increasingly hectic and still non-transparent consultation process. As to the requirements of transparency and copyright enforcement measures laid down in the AI Act and the related Directives on copyright, the third draft is toxic in its entirety – not only for authors in the text sector, but for the entire cultural and creative sources, the authors, artists, and performers, and their publishing or producing partners.

‍

No large language model without the work of professional human authors

As is well known, none of the existing large language models, on whose basis various chatbots are built, could function without the high-quality work of professional authors, journalists and literary translators. This was confirmed again by the scandalous AI policy recently put forward by Open A’s CEO Sam Altman: Without our work, the development of generative AI is terminated. On the other hand, the past two years have shown that AI applications are being misused to replace precisely those from which they had previously copied on a massive scale and in complete ignorance of the 3-step-test, which would also apply to TDM exceptions Art. 3 and 4 of the CDSM Directive (EU) 2019/790.

‍

Great imbalance in favour of the AI industry instead of a following EU law

We are dealing with the greatest imbalance since the very beginning of the digital age. The EU AI Act was designed to regulate this imbalance in such a way that, on the one hand, advanced technologies can be further developed – but without being developed at the detriment of authors, citizens, people whose works and data are the only relevant ingredient for any data-based system, commonly called ‘AI’.

This makes it even more necessary not to issue any further ‘free ride tickets’ within the Code of Practice, that allow profitable tech companies to exploit individual work and, in the case of book writers, of works created at private economic risk and at personal investment.

‍

The Code of Practice is not fit for Purpose

The Code of Practice is not fit for purpose and lacks respect on authors’ legit interests, hindering the enforcement of rights, and circumvent EU law and court rulings.

1. The third version of the Code of Practice is extraordinarily remote from reality when it comes to questions about material scraped from piracy sites. AI providers are politely requested to make “reasonable efforts” not to use pirated works. We wonder where this reluctance comes from, and even more, why only ‘commercially’ oriented piracy portals are mentioned –since most of them, and most often in the area of books or audiobooks, claim to have ‘non-commercial’ interests. This is an implicit and dangerous consent to use works taken from illegal portals that offer ‘free works’, and is not in line with EU law, as the European Commission stated clearly on 6^th March 2025[1] on the Meta case, who allegedly used the 7.5 million book works from the piracy site LibGen for their LLM[2], and not in line with CJEU rulings[3]. A TDM rights reservation is essential to justify the far-reaching and royalty-free exemption in favour of commercial users. The regulation must be interpreted by the CoP drafters in such a way that it does not cover works that have been made available illegally.
2. The third draft states: ‘Signatories are encouraged to make publicly available and keep up to date a summary of their copyright policy.’ This further weakening of obligations – from a detailed copyright policy to a summary that can be published or not, at the signatory’s convenience – will result in a lack of information necessary for authors to enforce their rights, such as updated names and languages of crawlers, or indications of which rights reservations are recognised and respected.
3. The third draft weakens obligations under the AI Act and EU copyright law, including but not limited to the measures related to compliance with rights reservations. The third draft still favours robots.txt protocols over other opt-outs despite repeated concrete examples and reasoned objections from authors’ associations and experts, and only cautiously asks providers to make “best efforts” to identify other opt-outs – and does not even oblige them to publish crawler names and their functions consistently. How exactly are rightsholders supposed to be able to declare their works with legible opt-outs? Robots.txt is neither suitable for every type of work nor accessible for every author as the primary rightsholder of TDM and AI rights. Furthermore, it is known that these opt-outs are also ignored.
4. The third draft favours unfounded privileges for small and medium-sized companies and downstream providers/operators that modify or fine-tune an existing, possibly non-EU AI model. We understand the need to support SMEs and smaller players – after all, the cultural economy is built on the shoulders of individuals, single authors, and thus self-employed people in the sense of micro-enterprises. However, we, as solo entrepreneurs are bound by the same basic legal standards as all other participants in civil society – and so should be all SME’s.
5. It is extremely disappointing and, in our opinion, not compatible with applicable EU law that the drafters propose not to keep any records or documentation during the content sets regarding the individual works. While authors are forced to provide each individual work, each article, each translation, each photo, with a title-specific reservation of rights, if they want to be protected from unfair exploitation, AI providers aren’t even supposed to bother to clarify rights for each work in content sets, repositories and corpora not acquired via the web. The fact that this is technically possible, and indeed almost trivially simple, is shown, among other authoritative sources, in the latest report by Prof. Dr. Sebastian Stober[4].
6. We are also dismayed to note that works and content sets that were exploited under the TDM exception in Art. 3 of the CDSM Directive (EU) 2019/790 and that are then further exploited in the numerous existing private partnerships between research entities and commercial and for-profit companies to develop their products, and in principle should guarantee authors the chance for an opt-out under Art. 4 (3), do not play a role.
7. Another peculiar case in the unfortunate combination of an opaque and IP-hostile process is the division of responsibilities between the template as laid down in the AI Act (AI Office) and the Code of Practice (working group on copyright and transparency). The most essential element of the AI Act for each author, namely a publicly accessible documentation of all title-specific works used, and sources including all time-sensitive data, within a template, is created and ‘negotiated’ in a flawed manner. The AI Act requires that the Summary must enable authors and further rightsholders to effectively exercise or enforce their rights. The draft template for the Summary, presented by the AI Office in January, did not meet this objective, as the drafters saw no need to make a title-specific documentation mandatory, which effectively makes any legal enforcement guaranteed under EU law impossible. It is to be hoped that the evaluation of the stakeholder survey presented on 21 March and the massive objections of the AI industry to basic transparency will not intimidate the AI Office and that it will remain true to the applicable laws.
8. The Code of Practice should also include a strict commitment to ensure that the TDM opt-out does not reduce the findability of content. The fact that the Code of Practice contains only minor requirements (‘…are encouraged to take appropriate measures…’) is fatal for journalistic and creative content, as users often only find media websites via search engines. Public service broadcasters e.g. in Germany have therefore decided to not declare a rights reservation via robots.txt, as this lowers their findability. If users are going to digest content via generative AI instead of direct visits via search engines, the traffic and advertising revenues on media websites continue to decline, with disastrous consequences for individual artists, journalists, and authors.

All these points will lead to legal uncertainty and to injustices and a lack of possibilities for legitimate law enforcement.

As currently drafted, the watering down and removal of obligations will lead to a dissolution of responsibility and will make the Code of Practice a template that will result in numerous legal disputes and CJEU interpretations.

The drafting of the Code of Practice was not collaborative with rightsholders at all.

The AI Office invited us, European authors and rightsholders, to participate in the drawing-up of this Code of Practice – a possibility expressly provided for in the AI Act itself, as we are most certainly “relevant stakeholders”. We welcomed this possibility. It was clear from the beginning that, while only GPAI providers would be the Code’s signatories, the relevance of the copyright-related obligations imposed by the AI Act on those providers makes the authors and further rightsholders’ community a natural, unavoidable beneficiary of said obligations whose expressly stated objective is to help us exercise and enforce our authors’ rights, copyright and related rights. Compliance with these obligations is what the GPAI Code of Practice is meant to support.

But in its current version, the draft is however completely unacceptable from the point of view of more than half a million individual authors CEATL, EFJ and EWC represent in the text area. It ignores the key concerns and recommendations we have detailed and reiterated in our comments on the previous iterations of the draft and in several joint letters sent. In doing so, the process makes a mockery of all our sustained and constructive efforts to make this Code of Practice fit for purpose and contributing to the proper application of the AI Act.

In the opening statement of the third draft, it is stated that

Like the first and second drafts, this document is the result of a collaborative effort involving hundreds of participants from across industry, academia, and civil society. It has been informed by three rounds of feedback, including on the previous two drafts, which has been insightful and instructive in our drafting process.»

We object.

Unless very substantial and substantive improvements are introduced, this Code of Practice will render the exercise and enforcement of our rights more difficult.

This is an outcome European authors and further rightsholders cannot accept nor help endorse. Therefore, unless said improvements are introduced, we submit that the AI Office ought to forfeit any claims that the Code of Practice’s final outcome is, in any manner, reflects a process whereby the concerns and recommendations of the authors rightsholders involved were also considered. Likewise, if such a scenario – which we hope can still be reversed – is confirmed, the active involvement of European authors ought under no circumstance be evoked by the AI Office to help render legitimacy to this Code of Practice.

Such claims would be wholly inaccurate and, in light of how the drafting process has been conducted thus far, may well be deemed inappropriate.

SIGNATORIES

European Council of Literary Translators’ Associations (CEATL)

European Federation of Journalists (EFJ)

European Writers’ Council (EWC)

ABOUT THE SIGNATORIES

CEATL (European Council of Literary Translators’ Associations) is an international non-profit organisation created in 1993 as a platform where literary translators’ associations from different European countries could exchange views and information and join forces to improve status and working conditions of translators. Today is the largest organisation of literary translators in Europe with 36 member associations from 28 countries, representing some 10,000 individual literary translators. www.ceatl.eu

EFJ (European Federation of Journalists) is the largest organisation of journalists in Europe, representing over 320,000 journalists in 73 journalists’ organisations across 45 countries. The EFJ is recognised by the European Union and the Council of Europe as the representative voice of journalists in Europe. The EFJ is a member of the European Trade Union Confederation (ETUC). www.europeanjournalists.com

EWC (European Writers’ Council) is the world’s largest federation representing authors from the book sector only and constituted by 50 national professional writers’ and literary translators’ associations from 32 countries. EWC members comprise over 220.000 professional authors, writing and publishing in 35 languages. www.europeanwriterscouncil.eu

PDF of Joint Letter EWC CEATL EFJ

[1] https://www.europarl.europa.eu/doceo/document/E-10-2025-000144-ASW_EN.html

[2] https://www.theatlantic.com/technology/archive/2025/03/libgen-meta-openai/682093/

[3] EuGH v. 10.4.2014 – C-435/12, Rn. 35 ff. – ACI Adam.

[4] https://papers.ssrn.com/sol3/papers.cfm?abstract_id=5165118

‍

‍

‍

‍

Original post:
Not fit for purpose: Writers, translators and journalists of the European text sector express strong opposition to the Third Draft of the EU’s Code of Practice under the AI Act’s implementation. - EWC - European Writers Council