In a statement made on 12 October 2023, the French collecting society Sacem, which represents most authors/composers and publishers of music in France, announced that it is opting out of machine learning training for the works in its repertoire. Sacem explains that it is basing its opt-out from generative AI systems on Article L122-5-3 of the French Intellectual Property Code (that implements Article 4(3) of Directive (EU) 2019/790), which allows rightholders to explicitly reserve the use of their works for text and data mining, including the reproductions necessary for the use of works to train generative AI models. Sacem justifies its decision within the current context of ‘the grandiose development of artificial intelligence (AI) tools’.
In practice, the implementation of such opt-out decisions is complicated, as at the moment there is no standardised way to opt out (see Paul Keller’s comments on the blog here). Sacem’s statement is nevertheless already symbolically significant, given the importance of the collecting society, not only for the music market (it regroups most music publishers, authors and composers in France), but also in the field of lobbying before the French and EU institutions (along with other European collecting societies).
Sacem is not only interested in communication and lobbying: in its statement it says that from now on data mining activities by entities developing artificial intelligence tools using works in the Sacem repertoire must be subject to prior authorisation in order to ensure fair remuneration for the authors, composers and music publishers it represents. In other words, Sacem does not intend to refuse such use of the works in its repertoire: by opting out, Sacem aims at (re)creating an exclusive right so that it can negotiate licensing agreements with the major players in the field of generative AI. The statement indeed explains that ‘entities that use Sacem’s works to supply their training bases and carry out data mining activities will have to request prior authorisation from Sacem and expressly negotiate the conditions of this exploitation.’
But what if certain authors or composers, members of Sacem, do not want to have their works used for data mining or AI generative activities? Will they be bound by the licences Sacem intends to negotiate with the major generative AI players? Under Sacem’s regulations its members, authors and publishers alike, have to grant Sacem performance and reproduction rights for all current and future works on an exclusive basis. By stating that it is ‘exercising its opt-out right’, Sacem seems to consider that the right to oppose under Article L122-5-3 of the French IPC (Article 4(3) CDSM) has already been assigned to it by its members, simply because they accepted its by-laws and regulations. However, this interpretation is far from obvious: Sacem may well have to obtain specific authorisation from the members that are interested in such licence agreements in order to negotiate on their behalf. Failing that, it would deny individual creators (members of Sacem) the right to opt out effectively for their works or even specific works, meaning that litigation between authors on one side and Sacem and its licensees on the other side cannot be ruled out.
________________________
To make sure you do not miss out on regular updates from the Kluwer Copyright Blog, please subscribe here.
Thank you very much, dear colleague, dear Brad! I completely agree with you:
My understanding is that authors who are members of SACEM have entrusted SACEM with the exclusive management of their reproduction and performance rights by virtue of their adherence to SACEM’s statutes and internal rules,- as you indicate. As such, SACEM must deny a third party the right to use the said works for data mining purposes (and must therefore be requested to do so by the said third party) because it is not itself expressly authorized by the said authors to grant such a right to a third party. In my opinion, the express agreement of the authors is therefore required in all cases for this new aspect of the reproduction and adaptation right.
One might even wonder whether data mining involves the moral rights of authors. In my opinion, it does.
I fully agree with your reference to moral rights. Moral rights under French law are not assignable and may not even be waived. Any agreement with AI operators should provide that the operators shall implement measures ensuring that authors’ right to attribution (citation) and to the integrity of the copyrighted work will be protected against violation. This implies that in the event that an outcome of the AI comes very close to a specific work used for training the AI and risks infringing it, the user should be advised of the risk by the operator, including the identification of the work. And the liability of the operator would be an issue under article 17 of the CSDM directive. Or the operator should have implemented algorisible ?thmic technical measures to prevent any outcome from coming so close to any specific copyrighted work that it is infringing. Would this be feasible ?