EU Parliament: a report on Generative Artificial Intelligence and copyright protection has been published.

The European Parliamentary Research Services (EPRS) have published a technical-legal report dedicated to copyright aspects in the field of generative artificial intelligence, highlighting the implications deriving from the non-transparent structure of the latest generation generative models and proposing regulatory and operational recommendations to ensure their legal compliance.

The document clarifies that the production of content by generative AI systems is based on a statistical link and not directly traceable with training data. Although it is not possible to deterministically reconstruct the correlation between the generated output and the original inputs, the probabilistic link between the two elements remains inherent in the very function of the generative model.

This feature raises two central issues from the point of view of copyright:

- Attribution: To what extent can it be determined whether content generated by an AI system is derived, in part or substantially, from protected works used in training the model?

- Originality and novelty: to what extent can an output be considered an original creation, and not a mere statistical reformulation - highly sophisticated but substantially reproductive - of pre-existing works?

The report highlights the absence of traceability mechanisms by design in the generative AI models currently in use, resulting in structural discontinuity between inputs (training data) and outputs (generated content). This absence undermines the possibility of verifying possible copyright infringements and limits the legal protection of derivative works, including in relation to fair remuneration systems for original authors.

In light of these critical issues, the report highlights the urgency of developing new regulatory models of documentation, accountability and transparency, capable of filling the current gaps.

To address the identified challenges in a systemic way, the European Parliamentary Research Services propose a series of operational and regulatory measures:

Developer empowerment: Entities designing and training generative AI systems should be required to proactively document the provenance of the data used and to enable transparent audit mechanisms on models.

Reform of remuneration mechanisms: fair compensation schemes should also extend to statistical use and indirect influence of legacy works, going beyond the traditional approach based solely on verbatim reproduction;

Interoperable standards and verifiable infrastructures: the adoption of open standards and the promotion of collaborative ecosystems between rights holders, developers, researchers and regulatory authorities are desirable.

The report concludes that the European Union is uniquely positioned to take the lead in developing technical and legal parameters to ensure transparency, attribution and accountability in the context of generative AI. The strengthening of copyright protection - also in a preventive key - is indicated as an essential condition for a sustainable technological transition that respects fundamental rights.