Joint letter to the European Parliament's JURI Committee on the upcoming own-initiative report on copyright and generative AI

Dear Member of the Legal Affairs (JURI) Committee, 

We are writing to you on behalf of a coalition of professional organisations representing the collective voice of hundreds of thousands of writers, translators, journalists, performers, composers, songwriters, screen directors, screenwriters, visual artists, and other artists and creative workers.                                                                     

First of all, we would like to thank the European Parliament and its Legal Affairs Committee for drafting a report on “Copyright and generative artificial intelligence – opportunities and challenges.” It is a crucial opportunity to engage in a long-overdue, democratic debate on generative AI and its impact on copyright – a debate that never took place when the text and data mining (TDM) exceptions (Articles 3 and 4 of the EU CDSM Directive) were adopted more than six years ago.                                                                                    

As of today, generative AI models have already exploited massive amounts of protected works without any authorisation, remuneration or transparency for the authors, artists and performers we represent. These generative AI models would not exist without the works created by our members: yet, they now directly compete with them, threatening to displace human creativity and labour with devastating economic effects on the cultural and creative sectors (CCS). We have never experienced copyright exceptions so unclear, so widely misused and so damaging to our creative communities. This is not just unfair – it is unacceptable.                                                                                   

We call on the Legal Affairs Committee to draft an ambitious report addressing our concerns around the implementation of the AI Act and the questionable applicability of the text and data mining (TDM) exception (Article 4 of the CDSM Directive). Ultimately, we urge MEPs to place the key principles of authorisation, remuneration and transparency at the heart of this report.

Our asks to Members of the European Parliament

In the context of the forthcoming European Parliament report on copyright and generative AI, we urge you to:

1.     Hold a democratic debate on the applicability of the TDM exceptions, clarifying their scope in a way that safeguards the legitimate interests of rightsholders and ensures compliance with the three-step test.[1]

The European Commission has retrospectively adopted a broad interpretation of Article 4 of the CDSM Directive to cover the systematic and extensive use of creators’ protected works and performances without any authorisation – despite the fact that Art. 4 of the CDSM Directive was adopted years before the sudden rise of generative AI technologies and the Directive does not mention or define “Artificial Intelligence” and “Generative AI”.

This interpretation does not meet any public policy objective and contradicts both international law and the CJEU case law, as the Court found that general opt-out mechanisms are not valid substitutes for consent and, by conditioning the exercise of exclusive rights to a formality, they violate Art. 5.2 of the Berne Convention (Soulier & Doke, 2016). It has also recently been called into question by several Member States, who expressed the view that copyright uses for AI training go beyond the scope of the TDM exception.[2] Last but not least, an EPRS study presented in the JURI Committee on 6 June found that the TDM exceptions were misread and not designed for this scale or purpose.   

Such interpretation has a devastating impact on authors and performers, but also fails to provide legal certainty to generative AI models, facing legal challenges and potential liability in the future. More generally, it has allowed generative AI models to put the cart before the horse by using this exception in full opacity without even giving a chance to creators to provide their consent and exercise their right of reservation.

In this context, various fundamental questions remain open, including (a) the type of uses covered by the TDM exceptions in the context of generative AI, (b) which rightsholders are entitled to reserve the rights (“opt out”) for different uses, and (c) whether rights reservations expressed by rightsholders after their works were scraped and used for training by AI providers can be enforced retroactively.

2.     Call for an effective and timely implementation of the AI Act and ensure a high level of transparency to protect authors’ and performers’ rights.

As a potential postponement of the application of the AI Act is considered, we reject such a delay and call instead for its effective and timely implementation addressing its shortcomings and ensuring that the GPAI Code of Practice and the transparency template do not undermine Union copyright law but instead allow the authors, performers and creative workers we represent to effectively exercise their rights.

As highlighted in a recent statement signed by a broad group of rightsholders’ organisations,[3] the third draft of Code of Practice undermines the AI Act’s obligation to ensure compliance with Union copyright law through language such as “reasonable efforts”, watering down GPAI providers’ responsibility and hindering compliance with rights reservations, as well as diluting other key provisions of Union copyright law and the AI Act itself. In this regard, Spain, Portugal, Italy and Hungary stressed the importance of safeguarding copyright and transparency under the AI Act,[4] supported by eleven additional Member States in the Education, Youth, Culture and Sport Council on 13 May.        

Likewise, the template summary of training data needs to lead to actionable transparency regarding all protected works and performances used for training. Its disclosure should not be hindered by trade secrecy claims. Disclosing the “ingredient list” (i.e. the training data) is essential for transparency and does not equate to disclosing the proprietary “recipe” used to create the models. In fact, withholding such information would disregard the legislative mandate of the AI Act, which states that the aim of the template should be to “facilitate parties with legitimate interests, including copyright holders, to exercise and enforce their rights under Union law” (Recital 107). Without adequate information about the training data, rightsholders will be unable to ascertain whether their works and performances have been used and will effectively be prevented from exercising their rights. 

Furthermore, we encourage the European Parliament to support a legal presumption of use of protected works by GPAI providers. In a context where those providers reject transparency and prevent rightsholders to prove that their works and performances have been used, such a presumption would ease the burden of proof currently placed on rightsholders and assist them in exercising their rights effectively.      

3.     Ensure authors and performers can effectively authorise the use(s) of their works and performances in the context of generative AI (opt-in) and encourage functional solutions to remunerate them in an appropriate and proportionate manner.

In our view, authors and performers should always be able to provide a prior explicit and informed authorisation for any use of their works and performances for the purpose of training generative AI. We reject the notion that such authorisation may be granted “on their behalf” by third entities unless such entitlement has been expressly and knowingly transferred to them.

Based on such authorisations, any licences should a) trigger the application of Articles 18 to 23 of the CDSM Directive, including the appropriate and proportionate remuneration of authors and performers, and b) remunerate for both the input and the output of GPAI models.

4.     Ensure that the moral rights and personal data of authors and performers are protected.

When generative AI technologies scrape and ingest the work of performers and other creative workers, this inevitably also involves the processing of their voice, likeness, and other personal data. The use of AI-generated deep fakes and other AI-manipulated content poses a significant threat, not only to our democracies and citizens’ trust in the authenticity of digital content but also to the reputation of our members. The personal data and moral rights of authors and performers are too often disregarded or ignored by generative AI models – those rights should be upheld and protected rather than ignored.  

To conclude, Europe’s creative communities call on you to draft an ambitious report addressing the several unresolved issues left by the current EU legal framework, while promoting the development of generative AI in full compliance with EU copyright law and the principles of informed authorisation, remuneration and transparency for authors, performers, and other rightsholders.

[1] The three-step test is a fundamental safeguard intended to strike a fair balance between rightsholders and content users by limiting copyright and neighbouring right exceptions to (1) certain special cases that (2) do not conflict with the normal exploitation of the works or other subject matter and (3) do not unreasonably prejudice the legitimate interests of rightsholders. First established by the Berne Convention, the test was included in Directive 2001/29 and in the 2019 CDSM Directive.
[2] https://data.consilium.europa.eu/doc/document/ST-16710-2024-REV-1/en/pdf
[3] https://europeanwriterscouncil.eu/wp-content/uploads/2025/04/Joint-statement-on-the-Third-Draft-Code-of-Practice-28-March-2025-updated-v-07042025-002.pdf
[4] https://data.consilium.europa.eu/doc/document/ST-8188-2025-REV-2/en/pdf