AI and Copyright: Key insights on the “publicly available detailed summary” requirement

The French High Council for Literary and Artistic Property (CSPLA) released its mission report[1] on December 11, 2024, concerning the implementation of the Artificial Intelligence Act (AIA) and its relationship with copyright compliance.

Transparency obligation under Article 53 of the AIA

Article 53 of the AIA establishes a transparency obligation that imposes distinct yet complementary requirements on AI system providers: the drafting of a sufficiently detailed summary (Art. 53(1)(d)) and the establishment of an internal compliance policy (Art. 53(1)(c)).

In this context, the CSPLA has been tasked with identifying the information that AI system providers must disclose, depending on the cultural sectors concerned, to ensure that authors and holders of related rights can effectively exercise their rights (“opt-out” mechanism).

Interdependence of transparency obligations

The report clarifies that the distinct obligations set out in Article 53 of the AIA are, in practice, inseparable. The publicly available summary must include the key elements of the internal compliance policy. However, the protection of trade secrets cannot justify withholding the list of protected elements used (though it may shield details on how they were used).

Proposed summary model

Pending the publication of a summary model by the European Commission’s AI Office, the CSPLA has put forward its own model (p. 30). Regarding copyright-protected content, the CSPLA model requires AI system providers to disclose information on:

• Harvested content, whether obtained from the internet directly or through an authorized third party;
• Datasets acquired from third parties;
• The use or non-use of prompts in training the AI system; and
• Synthetic data, meaning artificially generated data derived from human-created content.

AI system providers must include specific details on the methodology used to ensure compliance with EU law for each of these categories.

AI system providers training their models on copyright-protected content must integrate these organizational measures in advance, prior to the full entry into force of the AIA. While the CSPLA’s summary model is not legally binding, it serves as a practical reference point for initiating this sector-specific compliance process.

[1] Alexandra Bensamoun and Lionel Ferreira, « Rapport de la mission relative à la mise en œuvre du règlement européen établissant des règles harmonisées sur l’intelligence artificielle », presented to the CSPLA on December 9 and published on December 11, 2024.