MEPs attempt to strike a new balance between AI and copyright

One of the touchstone issues in the international debate about how to regulate AI is the relationship between AI technologies and copyright law. A recent study commissioned by the European Parliament’s Legal Affairs Committee examines how generative AI challenges core principles of EU copyright law. It points to a legal mismatch between AI training practices and current text and data mining exceptions, and the uncertain status of AI-generated content. 

This study is feeding in to the current draft report in the European Parliament led by rapporteur Axel Voss (European People’s Party, Germany) for the Legal Affairs Committee entitled “Copyright and Generative Artificial Intelligence – Opportunities and Challenges”. In an explanatory statement, Mr Voss writes that his report is “an attempt to bring a workable balance between new technology and copyright closer.” 

While the draft Voss report is non-legislative, it will provide a strong indication of position of the Parliament on this issue and include recommendations for future legislative proposals. 

Training AI systems 

A central concern is the use of copyrighted content in training Generative AI (GenAI) systems. The draft Voss report critiques the current reliance on Article 4 of the EU’s Digital Single Market Directive (“CDSM”, 2019/790), which introduced specific exceptions for text and data mining. These exceptions allow research organisations and certain other users to mine copyrighted content for non-commercial purposes in certain conditions. However, commercial uses generally require explicit permission, and rights holders may opt out by expressing their wishes. The scope and interpretation of these exceptions has become a highly disputed topic, especially as AI becomes more sophisticated and commercialised. 

In his draft report, Mr Voss argues that the text and data mining exceptions were not designed to regulate AI training and fail to adequately protect rights holders. The explanatory statement also says that “the lack of compensation results in enormous legal uncertainty.” 

To address this situation, the draft report proposes: 

• A dedicated legal framework for Generative AI training, either through a new exception or an expansion of Article 4; 
• A standardised, machine-readable opt-out mechanism for rights holders; and 
• A centralised registry, potentially managed by the EU Intellectual Property Office (EUIPO), to record opt-outs and facilitate licensing.

 According to the rapporteur, these measures aim to ensure that creators can control and be compensated for the use of their works in AI training. 

Transparency and enforcement

Transparency is a recurring theme in the draft report. It calls for an effective transparency requirement, on the basis that rights holders cannot know whether their content is being used to training AI systems. This transparency requirement would oblige AI developers to provide a comprehensive and detailed list of the protected content they have drawn on for this novel use.  

In this context, the draft report is very critical of the existing provision in the EU’s AI Act which mandates that providers of general-purpose AI models must make a "sufficiently detailed summary" of the content used for training their models publicly available. This summary should be based on a template recently provided by the European Commission’s AI Office. According to Mr Voss, this provision has so far been completely inadequate since it cannot provide clarity regarding the use of content precisely because it is a summary. 

Instead, the report calls for: 

• Detailed disclosure of all copyrighted content used in AI training and inference; 
• The establishment of an irrebuttable presumption of copyright infringement if transparency obligations are not met; and 
• The involvement of a trusted intermediary, such as the EUIPO, to notify rights holders and enforce compliance.

 According to the draft report, these provisions are designed to empower creators and ensure accountability among AI developers. 

GenAI outputs and copyright eligibility 

Another central legal question is whether works generated by AI can enjoy copyright protection and, if so, who should be regarded as the author. The draft report reaffirms that AI-generated content should not be eligible for copyright protection. It stresses the importance of distinguishing between human-created and AI-generated works, as this classification has significant legal implications. The draft report also warns against the risk of GenAI outputs infringing reproduction and communication rights, especially when such content competes directly with original works.  

Safeguarding the press  

The press sector receives special attention in the draft report, which emphasises the role of journalism in preserving democracy. According to the rapporteur, the processing of press content in an automated and generative manner must trigger a commensurate compensation. It is also important for press publishers that they remain identifiable to users, which may require an obligation to cite sources. At the same time, the draft report calls for safeguards to prevent GenAI systems from favouring certain publications, to the detriment of media pluralism.  

Future outlook

The draft report, which is still subject to change, concludes with a series of recommendations for the European Commission. These include an immediate assessment of the adequacy of current copyright laws and the development of legal and technical toolkits to raise awareness among AI developers. It also suggests that the EUIPO take on a central role in managing opt-outs and licensing, creating a “one-stop shop” for AI developers. 

Next steps

The deadline for amendments to the Axel Voss draft report is 12 September 2025. 

For further information contact Francine Cunningham