The European Parliament wants to force AI to disclose the works used for its training.

Since the rise of generative AI models, a debate has been taking place in the cultural industries: how to protect the works used to train these systems?

Indeed, books, images or music can be sucked up en masse by AI models, often without the creators being informed or compensated. Faced with these concerns, European institutions are gradually seeking to clarify the rules of the game…

A strong political signal for cultural industries

On March 10, 2026, the European Parliament adopted an own-initiative report on copyright and generative AI. Even if it doesn't immediately create new legal obligations, this text marks a clear political direction.

This means that AI systems operating in the European Union must respect copyright rules, regardless of where their training data was collected.

Adopted by 460 votes to 71, with 88 abstentions, the report aims to reiterate that European copyright law fully applies to generative AI models marketed within the Union.

For many publishers and cultural sector stakeholders, this vote represents political support against the practices of large technology companies. And for good reason: several professional organizations have been denouncing for months the massive use of copyrighted content to train artificial intelligence systems. According to them, generative models sometimes even rely on pirated databases, as was recently seen with Nvidia. Members of the European Parliament therefore want to strengthen transparency obligations, and AI providers should publish detailed summaries of the works used to train their models. This information should be standardized and machine-readable, to allow rights holders to more easily identify the use of their content. The report also emphasizes the need to protect the exclusive rights of creators. In other words, AI companies would not be able to automatically use copyrighted works, and a license or authorization would remain the rule when rights holders request it.

Towards a European Register of Works Used by AI

Among the most structuring proposals is the creation of a European register of works used to train AI models, in a register hosted by the EUIPO (European Union Intellectual Property Office).

The objective would be to identify the content used by AI systems and indicate whether its authors have authorized or refused this use. Companies would also have to disclose the websites or databases scraped to create their training datasets.

For Members of the European Parliament, failure to comply with these transparency obligations could be considered a violation of copyright. This approach aims to facilitate legal action for rights holders in an area where technical opacity makes litigation complex…